KR20240026931A - Professional signal profiler for base calling - Google Patents

Abstract

Start the system. The system includes memory and runtime logic. The memory stores multiple specialized signal profilers. Each expert signal profiler within the plurality of expert signal profilers is trained to maximize the signal-to-noise ratio of signals detected for an analyte in a particular analyte class and sequenced at a particular signal profile characterized in a particular training data set. The runtime logic that accesses the memory performs base calling operations by applying each specialized signal profiler in the plurality of specialized signal profilers to the sequenced signal in each detected signal profile for an analyte within each analyte class. It is configured to execute a calling operation.

Description

Professional signal profiler for base calling

priority application

This application is filed on June 13, 2022 and is entitled "Specialist Signal Profilers for Base Calling," U.S. Provisional Patent Application Serial No. 17/839,353 (Attorney Docket No. ILLM 1041-2/IP-2063-US) Claims the benefit of priority to U.S. Provisional Patent Application No. 63/223,408, filed July 19, 2021, entitled "Specialist Signal Profilers for Base Calling" (Attorney Docket No. ILLM 1041-1/IP) -2063-PRV). The priority applications are hereby incorporated by reference for all purposes.

Technology field

The disclosed technology relates to devices and corresponding methods for automatic analysis of images or recognition of patterns. Included herein are systems that transform an image for the purposes of (a) through (c): (a) enhancing its visual quality prior to recognition, (b) positioning the image relative to a sensor or stored prototype. (c) reducing the amount of image data by registering and registering, or discarding irrelevant data, and (c) measuring significant characteristics of the image. In particular, the disclosed technology relates to removing spatial crosstalk from sensor pixels using equalization-based image processing techniques.

References

The following is incorporated by reference for all purposes as if fully set forth herein:

U.S. Provisional Patent Application No. 17/308,035, entitled “Equalization-Based Image Processing and Spatial Crosstalk Attenuator,” filed May 4, 2021 (Attorney Docket No. ILLM 1032-2/IP-1991-PRV);

U.S. Provisional Patent Application No. 63/106,256, entitled “Systems and Methods for Per-Cluster Intensity Correction and Base Calling,” filed October 27, 2020;

U.S. Provisional Patent Application No. 15/909,437, entitled “Optical Distortion Correction for Imaged Samples,” filed March 1, 2018;

U.S. Provisional Patent Application No. 14/530,299, entitled “Image Analysis Useful for Patterned Objects,” filed October 31, 2014;

U.S. regular patent application Ser. No. 15/153,953, entitled “Methods and Systems for Analyzing Image Data,” filed December 3, 2014;

U.S. Provisional Patent Application No. 15/863,241, entitled “Phasing Correction,” filed January 5, 2018;

U.S. Provisional Patent Application No. 14/020,570, entitled “Centroid Markers for Image Analysis of High Density Clusters in Complex Polynucleotide Sequencing,” filed September 6, 2013;

U.S. Provisional Patent Application No. 12/565,341, entitled “Method and System for Determining the Accuracy of DNA Base Identifications,” filed September 23, 2009;

U.S. regular patent application Ser. No. 12/295,337, entitled “Systems and Devices for Sequence by Synthesis Analysis,” filed March 30, 2007;

U.S. Provisional Patent Application No. 12/020,739, entitled “Image Data Efficient Genetic Sequencing Method and System,” filed January 28, 2008;

U.S. Provisional Patent Application No. 13/833,619, entitled “Biosensors for Biological or Chemical Analysis and Systems and Methods for Same,” filed March 15, 2013 (Attorney Docket No. IP-0626-US);

U.S. Provisional Patent Application No. 15/175,489, entitled “Biosensors for Biological or Chemical Analysis and Methods of Manufacturing the Same,” filed June 7, 2016 (Attorney Docket No. IP-0689-US);

U.S. Provisional Patent Application No. 13/882,088, entitled “Microdevices and Biosensor Cartridges for Biological Or Chemical Analysis and Systems and Methods for the Same,” filed April 26, 2013 (Attorney Docket No. IP-0462-US);

U.S. Provisional Patent Application No. 13/624,200, entitled “Methods and Compositions for Nucleic Acid Sequencing,” filed September 21, 2012 (Attorney Docket No. IP-0538-US);

U.S. Provisional Patent Application No. 13/006,206, entitled “Data Processing System and Methods,” filed January 13, 2011;

U.S. Regular Patent Application No. 15/936,365, filed March 26, 2018 and entitled “Detection Apparatus Having a Microfluorometer, A Fluidic System, and a Flow Cell Latch Clamp Module”;

U.S. Provisional Patent Application No. 16/567,224, entitled “Flow Cells and Methods Related to Same,” filed September 11, 2019;

U.S. Provisional Patent Application No. 16/439,635, entitled “Device for Luminescent Imaging,” filed June 12, 2019;

U.S. Provisional Patent Application No. 15/594,413, entitled “Integrated Optoelectronic Read Head and Fluidic Cartridge Useful for Nucleic Acid Sequencing,” filed May 12, 2017;

U.S. Regular Patent Application No. 16/351, No. 193, entitled “Illumination for Fluorescence Imaging Using Objective Lens,” filed March 12, 2019;

U.S. Provisional Patent Application No. 12/638,770, entitled “Dynamic Autofocus Method and System for Assay Imager,” filed December 15, 2009;

U.S. Provisional Patent Application No. 13/783,043, entitled “Kinetic Exclusion Amplification of Nucleic Acid Libraries,” filed March 1, 2013; and

U.S. Provisional Patent Application No. 16/826,168, entitled “Artificial Intelligence-Based Sequencing,” filed March 21, 2020 (Attorney Docket No. ILLM 1008-20/IP-1752-PRV).

The subject matter discussed in this section should not be assumed to be prior art solely as a result of its mention in this section. Likewise, it should not be assumed that any matter related to the subject matter mentioned or provided as background in this section has been previously recognized in the prior art. The subject matter of this section merely represents various approaches and may itself constitute implementations of the claimed technology.

Base calling accuracy is important for high-throughput sequencing and downstream analyses, such as read mapping and genome assembly. This disclosure relates to optimizing image data for accurate base call clusters during a sequencing run. One problem in image data optimization is the variation in the intensity profile (or intensity distribution) of clusters in the base-called cluster population. This makes multi-cycle imaging of substrates (e.g. flow cells) with large numbers (e.g. thousands, millions, billions, etc.) of clusters difficult to manage, reducing data throughput and increasing error rates. Particularly harmful.

The intensity profiles of the millions of clusters on a flow cell can vary between individual clusters or between subpopulations of clusters. There are many potential reasons for this variation. This may result from differences in cluster brightness caused by fragment length distribution in the cluster population or unwanted light emissions from adjacent clusters (spatial crosstalk). It results from phase errors, which occur when a molecule in a cluster does not contain a nucleotide in some sequencing cycles and lags behind other molecules, or when a molecule contains more than one nucleotide in a single sequencing cycle. It can be. It may result from fading, i.e., an exponential decrease in the signal intensity of a cluster as a function of the number of sequencing cycles due to excessive cleaning and laser exposure as the sequencing run progresses. It may result from underdeveloped cluster colonies, i.e., small cluster sizes that result in empty or partially filled wells in the patterned flow cell. It may result from overlapping cluster colonies resulting from non-exclusive amplification. It may result from insufficient or uneven illumination, for example due to the cluster being located at the edge of the flow cell. It may originate from impurities (e.g. bubbles) in the flow cell that obfuscate the emitted signal. It may result from polyclonal clusters, i.e., when multiple clusters are deposited in the same well. This may result from different types of distortion in the image induced by the geometry of the optical lens. These distortions may include, for example, magnification distortion, skew distortion, translation distortion, and nonlinear distortions such as barrel distortion and pincushion distortion.

These variations can be corrected in a coarse-grained manner by training an intensity corrector on the entire population of clusters. Alternatively, one would train individual intensity correctors for each subpopulation of clusters, where the subpopulations are segmented in a way that minimizes sequencing errors and maximizes base call accuracy within the bounds of available computation. . The present invention relates to the latter, more detailed approach. More details are as follows.

In the drawings, like reference numbers generally refer to like parts throughout the different views. Additionally, the drawings are not necessarily to scale, with emphasis instead generally being placed upon illustrating the principles of the disclosed technology. In the following description, various implementations of the disclosed technology are described with reference to the drawings.
Figure 1 depicts an example sequencing environment with an imaging system.
2 is a block diagram illustrating an example two-channel, line-scanning modular optical imaging system that may be implemented in certain implementations.
Figure 3 shows image data subsets 1 to N generated for each class 1 to N of clusters located on the flow cell in respective spatial configurations 1 to N, trained to maximize the signal-to-noise ratio. An implementation example of each signal profiler 1 to N is shown.
4 illustrates an example configuration of a flow cell that can be imaged according to implementations disclosed herein.
Figure 5A shows the lanes of the upper surface of the flow cell.
Figure 5B shows swathes of tiles within a lane of the upper surface of the flow cell.
Figure 5C illustrates tiles within a swath of lanes of the upper surface of the flow cell.
Figure 5D shows subtiles within a tile within a swath of the upper surface of the flow cell.
Figure 6A shows one implementation of training respective surface-specific expert signal profilers for each cluster class during sequencing run 600.
Figure 6B shows one implementation of applying trained surface-specific expert signal profilers to image data subsets corresponding to each cluster class.
Figure 7A shows one implementation of training respective lane group-specific expert signal profilers for each cluster class.
Figure 7b shows one implementation of training respective lane-specific expert signal profilers for each cluster class.
Figure 7C shows one implementation of training respective swath-specific expert signal profilers for each cluster class.
Figure 7d shows one implementation of training respective tile-specific expert signal profilers for each cluster class.
Figure 7E shows one implementation of training respective subtile-specific expert signal profilers for each cluster class.
Figure 8 shows one implementation of applying trained lane group-specific expert signal profilers to image data subsets corresponding to individual cluster classes during a sequencing run.
Figure 9 shows one implementation of applying trained lane-specific expert signal profilers to image data subsets corresponding to individual cluster classes during a sequencing run.
Figure 10 shows one implementation of applying trained swath-specific expert signal profilers to image data subsets corresponding to each cluster class during a sequencing run.
Figure 11 shows one implementation of applying trained tile-specific expert signal profilers to image data subsets corresponding to individual cluster classes during a sequencing run.
Figure 12 shows one implementation of applying trained subtile-specific expert signal profilers to image data subsets corresponding to individual cluster classes during a sequencing run.
Figure 13 shows one implementation of specialized signal profilers for each/separate/different/independent sequencing cycles of each subseries of a sequencing run with a total of N sequencing cycles.
Figure 14 shows the results of each/separate/different/independent expert signal profilers for combinations of different spatial configurations (e.g. different subtiles) and different temporal configurations (e.g. different subseries of sequencing cycles). One implementation example is shown.
Figure 15 shows one implementation of each/separate/different/independent expert signal profilers for each cluster/well sequenced during a sequencing run.
16 shows one implementation of offline training of expert signal profilers on sequenced data from one or more completed/already run sequencing runs, and application of trained expert signal profilers on sequenced data from an ongoing sequencing run. An example is shown.
Figure 17 shows an example tile image divided into subtile images.
18 shows online training of expert signal profilers on sequenced data from early sequencing cycles of an ongoing sequencing run, and expert signal profilers trained on sequenced data from later sequencing cycles of an ongoing sequencing run. One implementation example of application of these is shown.
Figure 19 shows one implementation of training separate/separate/different/independent expert signal profilers for each signal distribution observed in sequenced data.
Figure 20 shows an example of signal distribution/signal profile/cluster intensity profile.
Figure 21 illustrates one implementation of a processing pipeline implementing the disclosed technology.
22A-22E illustrate equalizer coefficient sets of an example spatial equalizer.
Figures 23-31 illustrate one implementation of training an equalizer.
Figure 32A shows the base-by-base signal distribution of a population of clusters without the use of an equalizer and with a signal-to-noise ratio of 11.96 decibels (dB).
Figure 32b shows the signal distributions for each base of the same cluster group using the equalizer, and shows that the signal-to-noise ratio was improved to 13.13 dBs.
Figure 33 shows how the cost function for an expert signal profiler improves with each iteration of gradient descent.
Figure 34 is a plot showing the initial and final values of the cost function of Figure 33 as the expert signal profiler is adapted/trained/configured/updated at each sequencing cycle.
Figures 35A and 35B illustrate the improvement of primary analysis metrics for sequencing runs when adapting/training/configuring/updating an expert signal profiler.
Figures 36A and 36B show two plots evaluating the number of subtiles a sequencing tile can be divided into for adaptive equalization of each expert profiler.
Figure 37 depicts an example computer system that can be used to implement the disclosed techniques.

The following discussion is presented to enable any person skilled in the art to make and use the disclosed technology, and is presented in relation to specific applications and requirements thereof. Various modifications to the disclosed implementations will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other implementations and applications without departing from the spirit and scope of the disclosed technology. Accordingly, the disclosed technology is not intended to be limited to the illustrated implementations but is to be accorded the broadest scope consistent with the principles and features disclosed herein.

First, the signal profiler and then the disclosed professional signal profiler are described.

signal profiler

As used herein, a “signal profiler” maximizes the signal-to-noise ratio of a signal that is disturbed by noise. A signal profiler can be a value or function applied to data to modify it in a desired way. For example, data may be modified to increase its accuracy, relevance, or applicability to a particular situation. A signal profiler can be applied to data by any of a variety of mathematical operations, including but not limited to addition, subtraction, division, multiplication, or combinations thereof. Signal profilers can be mathematical expressions, logical functions, computer-implemented algorithms, etc. The data may be image data, electrical data, or a combination thereof.

In one implementation, the signal profiler is an equalizer (eg, spatial equalizer). Equalizers can be trained to maximize the signal-to-noise ratio of cluster intensity data in sequencing images (e.g., using least squares estimation, adaptive equalization algorithms). In some implementations, the equalizer is a LUT bank with a plurality of lookup tables (LUTs) with subpixel resolution, also referred to as “equalizer filters” or “convolution kernels.” In one implementation, the number of LUTs in the equalizer depends on the number of subpixels into which the pixels of the sequencing images can be divided. For example, if the pixels can be partitioned by nxn subpixels (eg, 5x5 subpixels), the equalizer generates n ² LUTs (eg, 25 LUTs).

In one implementation of a training equalizer, data from sequencing images are binned by well subpixel positions. For example, for a 5 x 5 LUT, approximately l/25 of the wells have their centers in bin(1,1) (e.g., top left corner of the sensor pixel), and l/25 of the wells have their centers in bin(1,1). 2), etc. In one implementation, the equalizer coefficients for each bin are determined using least squares estimation on the subset of data from the wells corresponding to each bin. In this way, the generated estimated equalizer coefficients are different for each bin.

Each LUT/equalizer filter/convolution kernel has a plurality of coefficients learned from training. In one implementation, the number of coefficients in the LUT corresponds to the number of pixels used to base call the cluster. For example, if the local grid of pixels (images or pixel patches) used to base call a cluster is of size pxp (e.g., a 9 x 9 pixel patch), then each LUT has p ² coefficients ( For example, 81 coefficients).

In one implementation, the training includes an equalizer configured to mix/combine the intensity values of pixels depicting intensity emissions from a base-called target cluster and intensity emissions from one or more adjacent clusters in a manner that maximizes the signal-to-noise ratio. Generate coefficients. Signal-to-noise In different lane groups, the maximized signal is the intensity emissions from the target cluster, and the noise minimized at the signal-to-noise ratio is the intensity emissions from adjacent clusters (e.g., to account for background intensity emissions), i.e., the spatial cross torque plus some random noise. The equalizer coefficients are used as weights, and mixing/combining involves performing an element-wise multiplication between the equalizer coefficients and the intensity values of the pixels to compute a weighted sum of the intensity values of the pixels, i.e. a convolution operation (i.e. , operations). Additionally, in some cases, the image data spans multiple color channels, and a set of equalizer coefficients is generated for each color channel (eg, one channel, three channels, four channels, etc.).

22A-22E illustrate equalizer coefficient sets of an example spatial equalizer. As shown by the heat maps, different sets of equalizer coefficients are configured to attenuate and enhance the signals of pixels differently depending on the positions of the pixels.

Figure 23 shows one implementation of training an equalizer. For the first sequencing cycle (Cycle 1), the equalizer of FIG. 23 has a first set of equalizer coefficients 2302 for the green color channel, and a second set of equalizer coefficients 2304 for the blue color channel. has Additionally, for the first sequencing cycle (Cycle 1), the first cluster (Cluster 1) has input image pixels 2306 for the green color channel and input image pixels 2308 for the blue color channel.

24, a first set of equalizer coefficients 2302 multiplies the input image pixels 2306 element-wise (2402) to produce a weighted sum 2316 for the green color channel. 25, a second set of equalizer coefficients 2304 multiplies the input image pixels 2308 element-wise (2502) to produce a weighted sum 2318 for the blue color channel.

Base call logic 2322 then predicts base call 2324 based on the weighted sums 2316 and 2318 using the expectation maximization (EM) algorithm discussed above. 26, based on predicted base calls 2324, the weighted sum 2316 for the green color channel is compared to the centroid value of paid bases 2612 for the green color channel with base call logic 2322. do. The comparison yields a base call error (2336) for the green channel. 26 , based on the predicted base call 2324, the weighted sum 2318 for the blue color channel is the centroid value of paid bases 2712 for the blue color channel with base call logic 2322. compared to The comparison yields a base calling error (2338) for the blue channel.

28 and 29, base call errors 2336, 2338 are used by update logic 2342 to update an updated first set of equalizer coefficients 2356 for the green color channel and the blue color channel. generates an updated second set of equalizer coefficients 2358.

In some implementations, the steps above are performed on multiple clusters. For example, for the three clusters shown in Figures 30 and 31, three updated versions of the first set of equalizer coefficients for the green color channel, and the first set of equalizer coefficients for the blue color channel. Two sets of three updated versions are produced. The three updated versions are: for the second sequencing cycle (Cycle 2), a first set of equalizer coefficients 2362 for the green color channel, and a first set of equalizer coefficients 2354 for the blue color channel. 2 is used to calculate the set.

Figure 32A shows the base-by-base signal distribution of a population of clusters without the use of an equalizer and with a signal-to-noise ratio of 11.96 decibels (dB). Figure 32b shows the signal distributions for each base of the same cluster group using the equalizer, and shows that the signal-to-noise ratio was improved to 13.13 dBs. The improvement in signal-to-noise ratio is visually observable by the earlier/more distinct base-specific clouds/distribution in Figure 32B compared to the base-specific clouds in Figure 32A.

Additional details regarding the equalizer may be found in U.S. Provisional Patent Application No. 17/, entitled “Equalization-Based Image Processing and Spatial Crosstalk Attenuator,” filed May 4, 2021, which is incorporated by reference as if fully set forth herein. 308,035 (Attorney Docket No. ILLM 1032-2/IP-1991-PRV).

Now, we discuss professional signal profilers.

Professional signal profiler

Global intensity correction applied at the pan-flow cell level or pan-sequencing run level does not take into account various noises in the image data. For example, non-linear distortion and noise can be induced by the shape of the optical lens that captures the image data. Additionally, imaged flow cells can also introduce distortions in the well pattern due to the manufacturing process (e.g., 3D bath effect introduced by joining or moving wells due to the non-rigidity of the substrate). Finally, the tilt of the flow cell within the holder is not taken into account by global intensity correction.

As used herein, a “professional signal profiler” is a signal profiler that is configured/trained to maximize the signal-to-noise ratio of a particular category/type/configuration/characteristic/class/bin of data. Discloses various professional signal profilers. For example, a “surface-specific expert signal profiler” is configured/trained to maximize the signal-to-noise ratio of sequencing data for clusters located on a specific surface or on a specific surface type/category/class (e.g., the top surface of a flow cell). or bottom surface or surface 1 to N). Similarly, “lane-specific expert signal profilers” are configured/trained to maximize the signal-to-noise ratio of sequencing data for clusters located in a particular lane or a particular lane type/category/class (e.g., central lane, peripheral lane, or flow Lanes 1 to N of cells). Additionally, “tile-specific expert signal profilers” are configured/trained to maximize the signal-to-noise ratio of sequencing data for a specific tile or a cluster located in a specific tile type/category/class (e.g., central tile, peripheral tile, or flow Tiles 1 to N of cells). Additionally, “subtile-specific expert signal profilers” are configured/trained to maximize the signal-to-noise ratio of the sequencing data of a specific subtile or clusters located in a specific subtile type/category/class (e.g. central subtile, Peripheral subtiles or subtiles 1 through N of the flow cell). More examples and details of the disclosed specialized signal profilers follow.

In some implementations, a single signal profiler may include a plurality of specific coefficient sets, such that each specific coefficient set is configured to maximize the signal-to-noise ratio of a specific category/type/configuration/characteristic/class/bin of data. /trained. In some implementations, a single signal profiler can include a variety of specific coefficient sets. For example, a “surface-specific expert coefficient set” is constructed/trained to maximize the signal-to-noise ratio of sequencing data for clusters located on a specific surface or on a specific surface type/category/class (e.g., the top surface of a flow cell or bottom surface or surfaces 1 to N). Likewise, a “lane-specific expert coefficient set” is constructed/trained to maximize the signal-to-noise ratio of sequencing data for clusters located in a particular lane or a particular lane type/category/class (e.g., central lane, peripheral lane, or flow cell). lanes 1 to N). Additionally, a “tile-specific expert coefficient set” is constructed/trained to maximize the signal-to-noise ratio of the sequencing data of a particular tile or a cluster located in a particular tile type/category/class (e.g., central tile, peripheral tile, or flow cell tiles 1 to N). Additionally, a “subtile-specific expert coefficient set” is constructed/trained to maximize the signal-to-noise ratio of the sequencing data of a particular subtile or clusters located in a particular subtile type/category/class (e.g. central subtile, peripheral Subtiles 1 to N of the subtile or flow cell). More examples and details of the specific coefficient sets disclosed follow.

The disclosed specialized signal profiler is applicable to clusters located on both patterned and unpatterned surfaces of the flow cell. Due to the unpatterned surface, clusters are randomly distributed on the flow cell. Randomly distributed clusters and their data (e.g., images) may be binned spatially, temporally, signally, or any combination thereof. Accordingly, expert signal profilers can be configured and trained on different configurations of differently binned randomly distributed clusters. By means of the patterned surface, clusters are positioned on the patterned wells in fixed positions. Patterned wells and constituent clusters can be binned spatially, temporally, signally, or any combination thereof. Accordingly, expert signal profilers can be configured and trained for different configurations of differently binned patterned clusters.

The disclosed expert signal profilers are configuration expert signal profilers trained to maximize the signal-to-noise ratio of image data generated for different configurations of a sequencing run. These configurations are spatial configurations associated with different regions on the flow cell, temporal configurations associated with different sequencing/imaging cycles of a sequencing run, and different distributions/patterns of signal profiles observed/encoded in the imaged data. It may be signal distribution configurations, or a combination thereof. Before describing various implementations of the systems and methods disclosed herein, it is useful to describe an example environment in which the techniques disclosed herein may be implemented.

Sequencing environment

1 depicts an example sequencing environment with imaging system 100. Exemplary imaging system 100 may include a device for acquiring or generating images of a sample. The example outlined in FIG. 1 illustrates an exemplary imaging configuration of a backlight design implementation. It should be noted that although systems and methods may sometimes be described herein in the context of example imaging system 100, these are merely examples of how implementations of the expert signal profiler disclosed herein may be implemented.

As can be seen in the example of Figure 1, a sample of interest is placed on a sample vessel 110 (e.g., a flow cell as described herein), which captures the sample under objective lens 142. It is located on stage 170. Light source 160 and associated optics direct a beam of light, such as laser light, to a selected sample location on sample vessel 110. The sample fluorescence and resulting light are collected by objective lens 142 and directed to an image sensor in camera system 140 to detect the fluorescence. Sample stage 170 is moved relative to objective lens 142 to position the next sample location on sample vessel 110 at the focus of objective lens 142. Movement of sample stage 110 relative to objective lens 142 may be accomplished by moving the sample stage itself, the objective lens, some other component of imaging system 100, or any combination of the foregoing. Additional implementations may also include moving the entire imaging system 100 relative to a stationary sample.

The fluid transfer module or device 100 directs the flow of reagents (e.g., fluorescently labeled nucleotides, buffers, enzymes, cleavage reagents, etc.) to (and through) the sample vessel 110 and waste valve 120. Sample container 110 may include one or more substrates on top of which a sample is provided. For example, for a system for analyzing multiple different nucleic acid sequences, sample vessel 110 may include one or more substrates to which the nucleic acids to be sequenced are bound, attached, or associated. In various embodiments, the substrate is any material to which nucleic acids can be attached, such as, for example, glass surfaces, plastic surfaces, latex, dextran, polystyrene surfaces, polypropylene surfaces, polyacrylamide gels, gold surfaces, and silicon wafers. It may comprise an inert substrate or matrix. In some applications, the substrate is within channels or other regions at multiple locations formed in a matrix or array throughout sample vessel 110.

In some implementations, sample vessel 110 may contain a biological sample that is imaged using one or more fluorescent dyes. For example, in certain embodiments, sample vessel 110 may be implemented as a patterned flow cell comprising a translucent cover plate, a substrate, and a liquid sandwiched between them, with the biological sample flowing inside the translucent cover plate. It may be located on the surface or on the inner surface of the substrate. A flow cell may include a large number (e.g., thousands, millions, or billions) of wells or regions that are patterned into an array (e.g., a hexagonal array, a rectangular array, etc.) defined within a substrate. Each region may form a cluster (e.g., a monoclonal cluster) of a biological sample such as DNA, RNA, or other genomic material that can be sequenced using, for example, synthetic sequencing. The flow cell can be further divided into multiple spaced lanes (e.g., eight lanes), each lane containing a hexagonal array of clusters. Exemplary flow cells that can be used in embodiments disclosed herein are described in U.S. Pat. No. 8,778,848.

The system also includes a temperature station actuator 130 and a heater/cooler 135 that can selectively adjust the temperature state of the fluid within the sample vessel 110. A camera system 140 may be included to monitor and track the sequencing of sample vessel 110. Camera system 140 may be implemented, for example, as a charge-coupled device (CCD) camera (e.g., a time delay integral (TDI) CCD camera), which includes various filters within filter switching assembly 145, objective lens 142 ), and can interact with the focusing laser/focusing laser assembly 150. Camera system 140 is not limited to CCD cameras, and other camera and image sensor technologies may be used. In certain implementations, the camera sensor may have a pixel size of about 5 to about 15 pm.

Output data from the sensors of camera system 140 can be used to analyze image data (e.g., image quality scoring) and characterize the laser beam (e.g., focus, shape, intensity, power, brightness, position) through a graphical user interface (e.g., image quality scoring). Reporting or display in a graphical user interface (GUI) and, as further described below, may be passed to a real-time analysis module (not shown), which may be implemented as a software application that dynamically corrects intensity noise in the image data.

Light source 160 (e.g., an excitation laser in an assembly optionally comprising multiple lasers) or another light source (optionally including one or more re-imaging lenses, fiber optic mounting, etc. can be included to illuminate the fluorescent sequencing reaction within the sample through illumination through an optical fiber interface. A low wattage lamp 165, focusing laser 150, and reverse dichroic 185 are also shown in the illustrated example. In some implementations, focusing laser 150 can be turned off during imaging. In other implementations, an alternative focusing configuration may include a second focusing camera (not shown), which includes a quadrant detector to measure the position of the scattered beam reflected from the surface simultaneously with data collection, a position-sensitive detector ( Position Sensitive Detector (PSD), or similar detector.

Although illustrated as a backlight device, other examples may include light from a laser or other light source directed onto the sample on sample vessel 110 through objective lens 142. Sample vessel 110 may ultimately be mounted on sample stage 170 to provide movement and alignment of sample vessel 110 relative to objective lens 142 . The sample stage may have one or more actuators that allow it to move in any of three dimensions. For example, in terms of a Cartesian coordinate system, actuators may be provided to enable the stage to move in the X, Y and Z directions relative to the objective lens. This can cause one or more sample locations on sample vessel 110 to be placed in optical alignment with objective lens 142 .

A focus (z-axis) component 175 is shown in this example as being included to control the positioning of the optical component relative to the sample vessel 110 in the focus direction (typically referred to as the z-axis or z-direction). It is done. Focus component 175 moves sample vessel 110 on sample stage 170 relative to an optical component (e.g., objective lens 142) to provide appropriate focus adjustment for imaging operations. It may include one or more actuators physically coupled to the sample stage, or both. For example, an actuator may be physically coupled to each stage, such as by mechanical, magnetic, fluidic or other attachment to the stage or direct or indirect contact with the stage. One or more actuators may be configured to move the sample stage in the z-direction while maintaining the sample stage in the same plane (e.g., maintaining a level or horizontal posture perpendicular to the optical axis). One or more actuators may also be configured to tilt the stage. For example, this can be done so that sample vessel 110 is dynamically leveled to account for any tilt of its surface.

Focusing of the system generally refers to aligning the focal plane of the objective lens with the sample to be imaged at a selected sample location. However, focusing can also refer to adjustments to the system to obtain desired characteristics for the representation of the sample, such as, for example, a desired level of sharpness or contrast for the image of the test sample. Because the usable depth of field of the objective lens' focal plane can be small (sometimes on the order of 1 pm or less), the focal component 175 closely follows the surface being imaged. Because the sample vessel is not perfectly flat as if fixed within the instrument, the focusing component 175 can be set to follow this profile while moving along the scanning direction (referred to herein as the y-axis).

Light emerging from the test sample at the sample location being imaged may be directed to one or more detectors of camera system 140. An aperture may be included and positioned to allow only light from the focus area to pass to the detector. The aperture may be included to improve image quality by filtering components of light coming from areas outside the focus area. An emission filter may be included in the filter switching assembly 145, which may be selected to record the determined emission wavelength and block any stray laser light.

Although not illustrated, a controller may be provided to control the operation of the scanning system. A controller may be implemented to control aspects of system operation such as focusing, stage movement, and imaging operations, for example. In various implementations, the controller may be implemented using hardware, algorithms (e.g., machine executable instructions), or a combination of the foregoing. For example, in some implementations, a controller may include one or more CPUs or processors with associated memory. As another example, a controller may include hardware or other circuitry for controlling the operation of a computer processor and a non-transitory computer-readable medium having machine-readable instructions stored thereon. For example, such circuits may include one or more of the following: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), programmable logic devices (PLDs), complex programmable logic devices (CPLDs), Programmable Logic Array (PLA), Programmable Array Logic (PAL) or other similar processing device or circuit. As another example, a controller may include a combination of such circuitry and one or more processors.

FIG. 2 is a block diagram illustrating an example two-channel, line-scanning modular optical imaging system 200 that may be implemented in certain implementations. Although systems and methods may sometimes be described herein in the context of example imaging system 200, it should be noted that these are merely examples of how implementations of the techniques disclosed herein may be implemented.

In some implementations, system 200 can be used for sequencing nucleic acids. Applicable techniques include those in which nucleic acids are attached at fixed locations within an array (e.g., wells of a flow cell) and the array is repeatedly imaged. In such implementations, system 200 can acquire images in two different color channels, which can be used to distinguish certain nucleotide base types from others. More specifically, system 200 may implement a process referred to as “base calling,” which generally refers to a base call (e.g., adenine (A), cytosine, etc.) for a given spot location in an image in an imaging cycle. refers to the process of determining thymine (C), guanine (G), or thymine (T)). During two-channel base calling, image data extracted from two images can be used to determine the presence of one of four base types by encoding the base identity as a combination of the intensities of the two images. For a given spot or position in each of the two images, the base identity is based on whether the combination of signal identities is [on, on], [on, off], [off, on], or [off, off]. can be decided.

Referring again to the imaging system 200, the system includes a line generation module (LGM) 210 with two light sources 211 and 212 disposed therein. The light sources 211 and 212 may be coherent light sources such as laser diodes that output a laser beam. The light source 211 may emit light of a first wavelength (eg, a red wavelength), and the light source 212 may emit light of a second wavelength (eg, a green wavelength). The light beam output from the laser light sources 211 and 212 may be directed through a beam shaping lens or lenses 213. In some implementations, a single light shaping lens can be used to shape the light beams output from both light sources. In other implementations, separate beam shaping lenses may be used for each light beam. In some examples, the beam shaping lens is a Powell lens, so that the light beam is shaped into a line pattern. The beam shaping lens of LGM 210 or other optical component of imaging system 200 may be configured to convert light emitted by light sources 211, 212 (e.g., one or more Powell lenses, or other beam shaping lenses, diffractive or scattering components). (by using) can be configured to be molded into a line pattern.

LGM 210 includes a mirror 214 and a semi-reflective mirror 215 configured to direct the light beam through a single interface port to an emission optics module (EOM) 230. More may be included. The light beam may pass through shutter element 216. EOM 230 may include an objective lens 235 and a z-stage 236 that longitudinally moves objective lens 235 closer to or further away from target 250 . For example, target 250 may include a liquid layer 252 and a translucent cover plate 251, such that the biological sample can be applied to the inner surface of the semitransparent cover plate as well as to the inner surface of a substrate layer located beneath the liquid layer. can be located The z-stage 236 may then move the objective lens to focus the light beam onto each inner surface of the flow cell (e.g., focused on the biological sample). The biological sample may be DNA, RNA, protein, or other biological material amenable to optical sequencing as known in the art.

The EOM 230 is configured to reflect the focus tracking light beam emitted from the focus tracking module (FTM) 240 onto the target 250 and then return the light from the target 250 back to the FTM (FTM). 240) may include a quasi-reflecting mirror 233 for reflecting inwards. FTM 240 may include a focus tracking optical sensor to detect characteristics of the returned focus tracking light beam and generate a feedback signal to optimize the focus of objective lens 235 on target 250.

EOM 230 may also include a quasi-reflective mirror 234 to direct light through objective 235 while allowing light returned from target 250 to pass. In some implementations, EOM 230 may include a tube lens 232. Light transmitted through the tube lens 232 may pass through the filter element 231 and to a camera module (CAM) 220. CAM 220 includes one or more optical sensors 221 for detecting light emitted from a biological sample in response to an incident light beam (e.g., fluorescence in response to red and green light received from light sources 211, 212). It can be included.

Output data from the sensors of CAM 220 may be transmitted to real-time analysis module 225. In various implementations, the real-time analysis module analyzes image data (e.g., image quality score, base calling, etc.) and displays beam characteristics (e.g., focus, shape, intensity, power, brightness, position), etc. in a graphical user interface (GUI). Executes computer-readable instructions, such as reporting or displaying These operations can be performed in real time during the imaging cycle, minimizing downstream analysis time and providing real-time feedback and troubleshooting during the imaging run. In implementations, the real-time analysis module may be a computing device (e.g., computing device 1000) that is communicatively coupled to and controls imaging system 200. In an implementation described further below, real-time analysis module 225 may further execute computer readable instructions to maximize the signal-to-noise ratio of output image data received from CAM 220.

Sequencing generates m sequencing images per sequencing cycle for the corresponding m image channels. That is, each of the sequencing images has one or more image (or intensity) channels (similar to the red, green, blue (RGB) channels of a color image). In one implementation, each image channel corresponds to one of a plurality of filter wavelength bands. In another implementation, each image channel corresponds to one of a plurality of imaging events in a sequencing cycle. In another implementation, each image channel corresponds to a combination of illumination using a specific laser and imaging through a specific optical filter. Image patches are tiled (or accessed) from each of the m image channels during a particular sequencing cycle. In different embodiments, such as 4-, 2-, and 1-channel chemistries, m is 4 or 2. In other embodiments, m is greater than 1, 3, or 4. In other implementations, the image may be in the blue and violet color channels instead of or in addition to the red and green channels.

Spatial configuration-specific expert signal profiler

Figure 3 shows image data subsets 1 to N generated for each class 1 to N of clusters located on the flow cell in respective spatial configurations 1 to N, trained to maximize the signal-to-noise ratio. An example implementation of each/separate/different/independent specialized signal profilers 1 to N is shown. Sequencing run 300 sequences a population of a plurality of sequencing cycles/imaging cycles 1 through K and generates image data depicting the intensity emissions of the population of clusters.

With respect to image data generated during sequencing run 300, the imaging cycle of the flow cell uses one or more coherent light sources to scan the flow cell area (e.g., using a line scanner) to collect image data for the entire flow cell. It is carried out so that it can be done. For example, imaging system 200, in conjunction with the system's optics, may use LGM 210 to line scan a flow cell with light having a wavelength within the red spectrum and to line scan a sample with light having a wavelength within the green spectrum. can do. In response to line scanning, the fluorescent dye located in different clusters of flow cell fluorescence and the resulting light can be collected by objective lens 235 and directed to the image sensor of CAM 220 for detection of the fluorescence. For example, the fluorescence of each cluster can be detected by several pixels of CAM 220. Subsequently, the image data output from the CAM 220 may be communicated to the real-time analysis module 225 for image noise correction and base calling.

A population of clusters is spatially distributed across the flow cell. The flow cell, and thus the population of clusters and image data, is partially segmented with different spatial configurations defined by different regions of the flow cell. So, for example, if the flow cell is partially in three rectangular regions, this represents three spatial configurations of the flow cell, three subpopulations or classes of clusters, three subsets of image data, and three specialized signal profilers. At the next detailed scale, each of the three rectangular regions of the flow cell can be further divided into three squares, creating a total of nine squares. This will result in 9 spatial configurations of the flow cell, 9 subpopulations or classes of classes, 9 subsets of image data, and 9 expert signal profilers.

Regarding segmentation of image data, the image data is divided into a plurality of image data subsets corresponding to each region of the flow cell. In various implementations, the size of the image data subset can be determined at, or on, the flow cell using the placement of fiducial markers or fiducials in the field of view of imaging system 200. The image data subsets may be partitioned such that the pixels of each image data subset have a predetermined number of reference points (e.g., at least 3 reference points, 4 reference points, 6 reference points, 8 reference points, etc.). For example, the total number of pixels in an image data subset can be predetermined based on predetermined pixel distances between boundaries and fiducials of the image data subset.

Additionally, the intensity data in each subset of image data may span multiple color channels (e.g., red, blue, green, and/or purple image channels). Accordingly, each expert signal profiler has a respective set of coefficients trained to maximize the signal-to-noise ratio of the intensity data in the corresponding color channel.

During inference, new/unseen/wild image data is partitioned with the same criteria as defined spatial configuration and used to generate each/separate/different/independent expert signal profiler in training. Each/separate/different/independent expert signal profiler is applied to the segmented new image data subset in such a way that the application of the expert signal profiler is limited to the corresponding new image data subset only.

Applying expert signal profilers 1 to N to image data subsets 1 to N generates image data subsets 1 to N with maximized signal-to-noise ratio, for further details see Filed May 4, 2021 It can be found in U.S. Provisional Patent Application Serial No. 17/308,035, entitled “Equalization-Based Image Processing and Spatial Crosstalk Attenuator” (Attorney Docket No. ILLM 1032-2/IP-1991-PRV). Other corrections for channel, spatial, or phase noise can be pre-applied to the raw image data, or subsequently maximize image data subsets 1 through N for signal-to-noise ratio.

The output of expert signal profilers 1 through N, i.e. the signal-to-noise ratio maximized image data subsets 1 through N, are provided as input to base caller 332 to make base calls 1 through N for the population of clusters. Create N. Base calling can be performed by fitting a mathematical model to the intensity data. Suitable mathematical models that can be used include, for example, k-means clustering algorithms, k-means pseudo-clustering algorithms, expectation maximization clustering algorithms, histogram-based methods, etc. Four Gaussian distributions can be fit to the set of two-channel intensity data such that one distribution applies for each of the four nucleotides represented in the data set. In one particular implementation, the expectation maximization (EM) algorithm may be applied. As a result of the EM algorithm, for each X, Y value (referring to each of the two channel intensities), a A value can be created. If the four bases give four distinct distributions, each X, Y intensity value will also have four associated likelihood values, one for each of the four bases. The maximum of the four possible values represents the base call. For example, if the cluster is “off” in both channels, the base call is G. If the cluster is "off" in one channel and "on" in the other, the base call is C or T (depending on which channel is on), and if the cluster is "on" in both channels, The base call is A.

FIG. 4 illustrates an example configuration of a flow cell 400 that can be imaged in accordance with implementations disclosed herein. In some implementations, flow cell 400 is patterned into a hexagonal array of aligned clusters or spots or features that can be imaged simultaneously during an imaging run. In other implementations, flow cell 400 may be patterned using a straight array, circular array, octagonal array, or some other array pattern. Flow cell 400 may have tens, hundreds, thousands, millions, or billions of clusters being imaged. In certain implementations, flow cell 400 may be patterned with millions or billions of wells divided into lanes. In this particular implementation, each well of flow cell 400 may include at least one cluster.

Surface-specific expert signal profiler

In some implementations, flow cell 400 may be a multi-plane sample containing multiple planes of clusters that are sampled during an imaging run. Examples of multiple planes of a cluster include an upper surface 402 and a lower surface 412. Accordingly, in one implementation, flow cell 400 may have two spatial configurations corresponding to upper and lower surfaces 402, 412, which in turn may result in two subpopulations or classes of clusters, image data. This results in two subsets of , and two specialized signal profilers.

Figure 6A shows one implementation of training each surface-specific expert signal profiler 604 for each cluster class 602. Each cluster class 602 includes groups of clusters each located on the upper surface 402, and clusters located on the lower surface 412. The result is a first expert signal profiler 604a configured to maximize the signal-to-noise ratio of the intensity data of clusters located on the upper surface 402, and the intensity data of the clusters located on the lower surface 412. A second specialized signal profiler 604b is configured to maximize the signal-to-noise ratio.

6B illustrates one implementation of applying trained surface-specific expert signal profilers 604 to image data subsets 632, 642 corresponding to each cluster class 602 during a sequencing run 600. An example is shown. In one implementation, flow cell 400 is imaged at the tile level. Accordingly, assuming top surface 402 has a first set of 1600 tiles and bottom surface 412 has a second set of 1600 tiles, first image data subset 632 has a first set of 1600 tiles. and a first set of -1600 tile images of the set, and the second image data subset 642 includes a second set of -1600 tile images of the second set of 1600 tiles.

The first expert signal profiler 604a maximizes the signal-to-noise ratio of the intensity data in the first image data subset 632 to generate a signal-to-noise ratio maximized version 634 of the first image data subset 632. It is configured to do so. The second expert signal profiler 604b maximizes the signal-to-noise ratio of the intensity data in the second image data subset 642 to generate a signal-to-noise ratio maximized version 644 of the second image data subset 642. It is configured to do so. Base caller 332 processes the signal-to-noise ratio maximized versions 634, 644 and generates base calls 638, 648.

As used herein, the term “maximized version” refers to data generated as output by a professional signal profiler. A maximized version of the input is the output produced by a professional signal profiler in response to processing the corresponding input. The maximized version of the input has a greater signal-to-noise ratio (SNR, S/R) than the corresponding input processed by a professional signal profiler. For example, a maximized version of the input (i.e., output) generated by an expert signal profiler is corrected by the expert signal profiler to have less spatial crosstalk and background noise compared to the corresponding input. Similarly, the maximized version of the input (i.e., output) generated by the expert signal profiler is corrected by the expert signal profiler to have less phasing and pre-phasing noise compared to the corresponding input.

Lane-specific expert signal profiler

In one implementation, the upper surface 402 may be divided or divided into a plurality of lanes 508a, 508b,...5081. In the example illustrated in Figure 5A, top surface 402 has eight lanes, but the number of lanes is implementation specific. Accordingly, in one implementation, flow cell 400 may have 16 spatial configurations corresponding to 8 lanes of upper surface 402 and 8 lanes of lower surface 412, resulting in 16 This results in 16 classes of subpopulations or clusters, 16 subsets of image data, and 16 expert signal profilers.

FIG. 7B shows one implementation of training each lane-specific expert signal profiler 714 for each cluster class 712. Each cluster class 712 includes groups of clusters respectively located on lanes 508a, 508b,..., 5081. The result is that the first expert signal profiler 714a is configured to maximize the signal-to-noise ratio of the intensity data of the cluster located in the first lane 508a, and maximize the signal-to-noise ratio of the intensity data of the cluster located in the second lane 508b. and a second expert signal profiler 714b configured to do so (continued with a first expert signal profiler 7141 configured to maximize the signal-to-noise ratio of the intensity data of the cluster located in the first lane 5081).

9 shows lane-specific expert signal profilers 714 trained on image data subsets 902, 912, ..., 922 corresponding to each cluster class 712 during a sequencing run 900. ) shows an implementation example of applying. In one implementation, flow cell 400 is imaged at the tile level. Accordingly, assuming each lane has 200 tiles, the first image data subset 902 includes a first set of 200 tile images of the 200 tiles on the first lane 508a, the second Image data subset 912 includes a second set of a second set of 200 tiles on second lane 508b, and so on (continuing with first image data subset 922).

The first expert signal profiler 714a maximizes the signal-to-noise ratio of the intensity data in the first image data subset 902 to generate a signal-to-noise ratio maximized version 904 of the first image data subset 902. It is configured to do so. The second expert signal profiler 714b maximizes the signal-to-noise ratio of the intensity data in the second image data subset 912 to generate a signal-to-noise ratio maximized version 914 of the second image data subset 912. (Continuing with signal-to-noise ratio maximizing version 924 of first image data subset 922). Base caller 332 processes signal-to-noise ratio maximizing versions 904, 914,..., 924 and generates base calls 908, 918,..., 928.

Lane group-specific expert signal profiler

In some implementations, lanes may be grouped into lane groups 502a, 502b, and 502c. Examples of lane groups include upper perimeter lanes, central lanes, and lower perimeter lanes. Accordingly, in one implementation, flow cell 400 may have six spatial configurations corresponding to three lane groups on the upper surface 402 and three lane groups on the lower surface 412, which This results in 6 subpopulations or classes of clusters, 6 subsets of image data, and 6 expert signal profilers.

FIG. 7A shows one implementation of training respective lane group-specific expert signal profilers 704 for each cluster class 702. Each cluster class 702 includes groups of clusters located respectively on lane groups 502a, 502b, and 502c. The result is a first expert signal profiler 704a configured to maximize the signal-to-noise ratio of the intensity data of the clusters located on the first lane group 502a, the intensity data of the clusters located on the second lane group 502b. a second specialized signal profiler 704b configured to maximize the signal-to-noise ratio of the intensity data of the clusters located on the third lane group 502c, and a third specialized signal profiler 704c configured to maximize the signal-to-noise ratio of the intensity data of the clusters located on the third lane group 502c. )am.

8 shows the application of trained lane group-specific expert signal profilers 704 to image data subsets 802, 812, 822 corresponding to each cluster class 702 during a sequencing run 800. Shows an implementation example of what is done. In one implementation, flow cell 400 is imaged at the tile level. Therefore, assuming the first lane group 502a has 600 tiles, the second lane group 502b has 600 tiles, the third lane group 502c has 400 tiles, and the first image Data subset 802 includes a first set of 600 tile images of the 600 tiles on a first lane group 502a, and a second image data subset 812 includes 600 tile images on a second lane group 502b. and a second set of 600 tile images of tiles, and the third image data subset 822 includes a third set of 400 tile images of 400 tiles on the third lane group 502c.

The first expert signal profiler 704a maximizes the signal-to-noise ratio of the intensity data in the first image data subset 802 to generate a signal-to-noise ratio maximized version 804 of the first image data subset 802. It is configured to do so. The second expert signal profiler 704b maximizes the signal-to-noise ratio of the intensity data in the second image data subset 812 to generate a signal-to-noise ratio maximized version 814 of the second image data subset 812. It is configured to do so. A third expert signal profiler 704c maximizes the signal-to-noise ratio of the intensity data in the third image data subset 822 to generate a signal-to-noise ratio maximized version 824 of the third image data subset 822. It is configured to do so. Base caller 332 processes signal-to-noise ratio maximizing versions 804, 814, 824 and generates base calls 808, 818, 828.

Swath-specific expert signal profiler

In some implementations, each lane includes one or more columns/swaths of tiles 506a, 506b, as shown in Figure 5B. Accordingly, in one implementation, flow cell 400 may have 32 spatial configurations corresponding to 32 swaths of tiles of upper surface 402 and 32 swaths of tiles of lower surface 412. There are, which results in 32 subpopulations or classes of clusters, 32 subsets of image data, and 32 expert signal profilers.

FIG. 7C shows one implementation of training respective swath-specific expert signal profilers 724 for each cluster class 722. Each cluster class 722 includes groups of clusters located respectively on swaths 506a and 506b. The result is a first expert signal profiler 724a configured to maximize the signal-to-noise ratio of the intensity data of the clusters located on the first swath 506a, and the clusters located on the second swath 506b. A second professional signal profiler 724b is configured to maximize the signal-to-noise ratio of the intensity data.

10 shows the application of trained swath-specific expert signal profilers 724 to image data subsets 1002, 1012 corresponding to respective cluster classes 722 during a sequencing run 1000. An implementation example is shown. In one implementation, flow cell 400 is imaged at the tile level. Thus, assuming first swath 506a has 100 tiles and second swath 506b has 100 tiles, first image data subset 1002 is Containing a first set of 100 tile images of 100 tiles, second image data subset 1012 includes a second set of 100 tile images of 100 tiles on second swath 506b.

The first expert signal profiler 724a maximizes the signal-to-noise ratio of the intensity data in the first image data subset 1002 to generate a signal-to-noise ratio maximized version 1004 of the first image data subset 1002. It is configured to do so. The second expert signal profiler 724b maximizes the signal-to-noise ratio of the intensity data in the second image data subset 1012 to generate a signal-to-noise ratio maximized version 1014 of the second image data subset 1012. It is configured to do so. Base caller 332 processes signal-to-noise ratio maximizing versions 1004, 1014, 1024 and generates base calls 1008, 1018.

Tile-specific expert signal profiler

Each swath includes a plurality of tiles 512a, 512b,..., 512t, as shown in FIG. 5B. The number of tiles in each swath is implementation specific, and in different examples there may be 50 tiles, 60 tiles, 80 tiles, etc. For example, consider that each swath has 100 tiles. Flow cell 400 will then have 200 tiles per lane, producing 1600 tiles for the upper surface 402 and another 1600 tiles for the lower surface 412 (i.e., 3200 tiles total) field). Accordingly, in one implementation, flow cell 400 may have 3200 spatial configurations corresponding to 3200 tiles, which results in 3200 subpopulations or classes of clusters, 3200 subsets of image data. , and 3200 professional signal profilers.

FIG. 7D shows one implementation of training respective tile-specific expert signal profilers 734 for each cluster class 732. Each cluster class 732 includes groups of clusters respectively located on tiles 512a, 512b,..., 512t. The result is a first expert signal profiler 734a configured to maximize the signal-to-noise ratio of the intensity data of the cluster located in the first tile 512a, and maximize the signal-to-noise ratio of the intensity data of the cluster located in the second tile 512b. a second expert signal profiler 734b configured to do so (continued with a t-th expert signal profiler 734t configured to maximize the signal-to-noise ratio of the intensity data of the cluster located in the t-th tile 512t);

11 shows one implementation of applying trained tile-specific expert signal profilers 734 to image data subsets 1102, 1112, ..., 1122 corresponding to each cluster class 732. (during sequencing run 1100). In one implementation, flow cell 400 is imaged at the tile level. Accordingly, for tiles 512a, 512b,..., 512t, the first image data subset 1102 includes the first tile image of the first tile 512a, and the second image data subset ( 1112 includes a second tile image of second tile 512b (continued with t-th image data subset 1122 including a t-th tile image of t-tile 512t (shown in FIG. 5C) ).

The first expert signal profiler 734a maximizes the signal-to-noise ratio of the intensity data in the first image data subset 1102 to generate a signal-to-noise ratio maximized version 1104 of the first image data subset 1102. It is configured to do so. The second expert signal profiler 734b maximizes the signal-to-noise ratio of the intensity data in the second image data subset 1112 to generate a signal-to-noise ratio maximized version 1114 of the second image data subset 1112. (Continuing with the signal-to-noise ratio maximizing version 1124 of the t image data subset 1122). Base caller 332 processes signal-to-noise ratio maximizing versions 1104, 1114,..., 1124 and generates base calls 1108, 1118,..., 1128.

Subtile-specific expert signal profiler

Each tile may be divided into: a plurality of subtiles 518a, 518b and 518s, as shown in Figure 5D. The number of subtiles a tile can be implementation specific may be, in different examples, 10 subtiles, 30 subtiles, 50 subtiles, etc. For example, consider that each tile is divided into 9 subtiles. Flow cell 400 will then have a total of 28,800 subtiles for 3200 tiles. Thus, in one implementation, flow cell 400 may have 28800 spatial configurations corresponding to 28800 tiles, which results in 28800 subpopulations or classes of clusters, 28800 subsets of image data. , resulting in 28800 professional signal profilers.

Figure 7E shows one implementation of training each subtile-specific expert signal profiler 744 for each cluster class 742. Each cluster class 742 includes groups of clusters each located on subtiles 518a, 518b, 518c, 518d,..., 518 s. The result is a first expert signal profiler 744a configured to maximize the signal-to-noise ratio of the intensity data of the cluster located in the first subtile 518a, and the signal-to-noise ratio of the intensity data of the cluster located in the second subtile 518b. A second specialized signal profiler 744b configured to maximize, a third specialized signal profiler 744c configured to maximize the signal-to-noise ratio of the intensity data of the cluster located in the third subtile 518c, and a fourth subtile ( a fourth professional signal profiler 744d configured to maximize the signal-to-noise ratio of the intensity data of the cluster located in 518d), and the like (a fourth specialized signal profiler 744d configured to maximize the signal-to-noise ratio of the intensity data of the cluster located in the s subtile 518s). s continued as a professional signal profiler (744s).

12 shows the subtile-specific full text trained on image data subsets 1202, 1212, 1222, 1232,..., 1242 corresponding to each cluster class 742 during sequencing run 1200. One implementation of applying signal profilers 744 is shown. In one implementation, flow cell 400 is imaged at the subtile level. Accordingly, for 518a, 518b, 518c, 518d,..., 518s, the first image data subset 1202 includes the first subtile image patch of the first subtile 518a, and the second image data Subset 1212 includes a second subtile image patch of second subtile 518b, and third image data subset 1222 includes a third subtile image patch of third subtile 518c. And, the fourth image data subset 1232 includes the fourth subtile image patch of the fourth subtile 518d (the sth image including the sth subtile image patch of the sth subtile 518s). Continued with Data Subsets (1242).

The first expert signal profiler 744a maximizes the signal-to-noise ratio of the intensity data in the first image data subset 1202 to generate a signal-to-noise ratio maximized version 1204 of the first image data subset 1202. It is configured to do so. The second expert signal profiler 744b maximizes the signal-to-noise ratio of the intensity data in the second image data subset 1212 to generate a signal-to-noise ratio maximized version 1214 of the second image data subset 1212. It is configured to do so. A third expert signal profiler 744c maximizes the signal-to-noise ratio of the intensity data in the third image data subset 1222 to generate a signal-to-noise ratio maximized version 1224 of the third image data subset 1222. It is configured to do so. A fourth expert signal profiler 744d maximizes the signal-to-noise ratio of the intensity data in the fourth image data subset 1232 to generate a signal-to-noise ratio maximized version 1234 of the fourth image data subset 1232. (Continuing with the signal-to-noise ratio maximizing version 1244 of the first image data subset 1242). Base caller 332 processes the signal-to-noise ratio maximizing versions 1204, 1214, 1224, 1234,..., 1244 and generates base calls (base calls 1208, 1218,

1228, 1238, ..., 1248).

Temporal configuration-specific expert signal profiler

Figure 13 shows an example implementation of each/separate/different/independent expert signal profilers for the sequencing cycles of each subseries of sequencing run 1300 with a total of N sequencing cycles. The first expert signal profiler 1312 is configured to maximize the signal-to-noise ratio of the intensity data of clusters located on subtile M and generated during sequencing cycles 1 to N1. The second expert signal profiler 1314 is configured to maximize the signal-to-noise ratio of the intensity data of clusters located on subtile M and generated during sequencing cycles N1 + 1 to N2. A third expert signal profiler 1318 is configured to maximize the signal-to-noise ratio of the intensity data of clusters located on subtile M, generated during sequencing cycles N2 + 1 to N. Other examples of temporal configurations include sequencing cycles of a first read of a sequencing run (Read 1), and sequencing cycles of a second read of a sequencing run (Read 2).

Figure 14 shows the results of each/separate/different/independent expert signal profilers for combinations of different spatial configurations (e.g. different subtiles) and different temporal configurations (e.g. different subseries of sequencing cycles). One implementation example is shown. In one implementation, cluster classes 1410 are defined by different spatial configurations (e.g., different subtiles). In one implementation, cluster subclasses 1412, 1414, 14 are defined by different temporal configurations (eg, different subseries of sequencing cycles).

The first expert signal profiler 1422 is configured to maximize the signal-to-noise ratio of the intensity data of clusters located on subtile 518a and generated during sequencing cycles 1 through N1. The second expert signal profiler 1424 is configured to maximize the signal-to-noise ratio of the intensity data of the clusters located on subtile 518a, generated during sequencing cycles N1 + 1 through N2. The third expert signal profiler 1428 is configured to maximize the signal-to-noise ratio of the intensity data of the clusters located on subtile 518a and generated during sequencing cycles N2 + 1 to N.

The fourth expert signal profiler 1432 is configured to maximize the signal-to-noise ratio of the intensity data of clusters located on subtile 518b and generated during sequencing cycles 1 through N1. The fifth expert signal profiler 1434 is configured to maximize the signal-to-noise ratio of the intensity data of clusters located on subtile 518b and generated during sequencing cycles N1 + 1 to N2. The sixth expert signal profiler 1438 is configured to maximize the signal-to-noise ratio of the intensity data of clusters located on subtile 518b and generated during sequencing cycles N2 + 1 to N.

Cluster/well-specific expert signal profiler

Figure 15 shows one implementation of each/separate/different/independent signal profilers for each cluster/well sequenced during a sequencing run. Clusters/wells on the flow cell can be pre-identified by their position coordinates. These position coordinates of clusters/wells train cluster/well expert signal profilers during training on cluster/well intensity data and during inference on cluster/well intensity data inference based on the position coordinates of clusters/wells. It can be used to apply professional signal profilers for each cluster. In Figure 15, cluster/well population 1502 has N clusters/well, so expert signal profilers 1508 includes each of N per-cluster/well expert signal profilers. In other implementations, as discussed above, different cluster/well signal profilers may also be trained for different temporal configurations.

offline training

16 shows one implementation of offline training of expert signal profilers on sequenced data from one or more completed/already run sequencing runs, and application of trained expert signal profilers on sequenced data from an ongoing sequencing run. An example is shown. During training stage 1602, training data 1612 is generated. Training data 1612 includes sequenced data from one or more completed/already executed sequencing runs.

Segmentation logic 1622 segments training data 1612 based on one or more configurations selected from different spatial configurations, temporal configurations, signal distribution configurations, or any combination thereof. The results are segmented training data 1632 with configuration-specific training data subsets 1 through N. For example, training data 1612 may include K images of a tile from K imaging cycles of a completed/already executed sequencing run, where each tile image has multiple color channels. Figure 17 shows an example of a tile image with C color channels. In this case, segmentation logic 1622 logically partitions each tile image of training data 1612 into subtile images by specifying pixel ranges. For example, a first subtile image of a tile may range from pixels 1 to 500, and a second subtile image of a tile may range from pixels 501 to 1000. Pixel ranges can be defined using fiducial markers as discussed above.

Offline training logic 1642 trains each/separate/different/independent expert signal profiler 1 to N for each configuration-specific training data subset 1 to N. The result is trained expert signal profilers 1 to N. Returning to the example of Figure 17, the expert signal profiler is trained to maximize the signal-to-noise ratio of each subtile image in training data 1612.

Inference data 1618 is generated during inference stage 1608. Inference data 1618 includes sequenced data from an ongoing sequencing run (e.g., the first i cycle of an ongoing sequencing run).

Segmentation logic 1622 segments inference data 1618 based on the same one or more configurations that were used to segment training data 1612 during training stage 1602. The result is segmented inference data 1638 with configuration-specific inference data subsets 1 through N. For example, inference data 1618 may include K images of a tile from K imaging cycles of an ongoing sequencing run, with each tile image having multiple color channels. Returning to the example of FIG. 17 , segmentation logic 1622 logically partitions each tile image of inference data 1618 into subtile images by specifying identical pixel ranges used to segment training data 1612. .

Runtime logic 1648 applies each of the trained expert signal profilers 1 through N 1658 to each configuration-specific inference data subsets 1 through N. Returning to the example of Figure 17, the corresponding trained expert signal profiler is applied to maximize the signal-to-noise ratio of each subtile image in the segmented inferred data 1638.

online training

18 shows online training of expert signal profilers on sequenced data from early sequencing cycles of an ongoing sequencing run, and expert signal profilers trained on sequenced data from later sequencing cycles of an ongoing sequencing run. One implementation example of application of these is shown. Inference data 1802 is generated during the inference stage 1802. Inferred data 1812 includes sequenced data from previous sequencing cycles (e.g., Cycles 1 through N1) of the ongoing sequencing run.

Segmentation logic 1622 segments inference data 1812 based on one or more configurations selected from different spatial configurations, temporal configurations, signal distribution configurations, or any combination thereof. The result is segmented inference data 1832 with configuration-specific training data subsets 1 to N. For example, inference data 1812 may include N1 images of tiles from N1 initial sequencing cycles of an ongoing sequencing run, with each tile image having multiple color channels. Figure 17 shows an example of a tile image with C color channels. In this case, segmentation logic 1622 logically partitions each tile image of inferred data 1812 into subtile images by specifying pixel ranges. For example, a first subtile image of a tile may range from pixels 1 to 500, and a second subtile image of a tile may range from pixels 501 to 1000. Pixel ranges can be defined using fiducial markers as discussed above.

Offline training logic 1842 trains each/separate/different/independent expert signal profiler 1 to N for each configuration-specific training data subset 1 to N. The result is trained expert signal profilers 1 to N. Returning to the example of Figure 17, the expert signal profiler is trained to maximize the signal-to-noise ratio of each subtile image in the inferred data 1812.

Inference data 1818 is also generated during the inference stage 1802. Inferred data 1818 includes sequenced data from later sequencing cycles (e.g., cycles N1 +1 to N2) of an ongoing sequencing run.

Segmentation logic 1622 may segment inference data 1818 based on the same one or more configurations that were used to segment inference data 1812 during an earlier sequencing cycle (e.g., Cycles 1 through N1) of an ongoing sequencing run. segment. The result is segmented inference data 1838 with configuration-specific inference data subsets 1 through N. For example, inference data 1818 may include N2 images of tiles from N2 later sequencing cycles of an ongoing sequencing run, with each tile image having multiple color channels. Returning to the example of Figure 17, segmentation logic 1622 logically partitions each tile image of speculation data 1818 into subtile images by specifying identical pixel ranges used to segment speculation data 1812. .

Runtime logic 1648 applies each of the trained expert signal profilers 1 through N 1858 to each configuration-specific inference data subsets 1 through N. Returning to the example of Figure 17, the corresponding trained expert signal profiler is applied to maximize the signal-to-noise ratio of each subtile image in the segmented inferred data 1838.

In some implementations, the training process is repeated iteratively, with each trained expert signal profiler 1 through N 1858 completing a subsequent sequencing cycle (e.g., cycles N1+1 through N2) of the ongoing sequencing run. is retrained/further trained on the segmented inference data 1838 and applied to the segmented inference data from later sequencing cycles (e.g., N2+1 to N3 cycles) of the ongoing sequencing run.

The control logic (not shown) may repeat in each successive sequencing cycle of an ongoing sequencing run or in successive subseries sequencing cycles of an ongoing sequencing run (e.g., every 10 or 20 sequencing cycles). : (i) segment the current batch of image data based on one or more configurations selected from different spatial configurations, temporal configurations, signal distribution configurations, or a combination thereof, and (ii) for each segmented current batch of image data retrain the trained expert signal profilers 1 to N of, (iii) segment the next batch of image data by the same criteria as the current batch of image data, and (iv) retrain each of the retrained expert signal profilers 1 to N It is applied to the next batch of segmented image data.

Signal Distribution Configuration-Specific Professional Signal Profiler

Figure 19 shows one implementation of training separate/separate/different/independent expert signal profilers for each signal distribution observed in sequenced data. In some implementations, each signal distribution can be observed in sequenced data from an offline/already run sequencing run. In other embodiments, the respective signal distribution can also be observed in sequenced data from an online/ongoing sequencing run (e.g., observed in the first 10 sequencing cycles of an ongoing sequencing run).

Figure 20 shows an example of signal distribution/signal profile/cluster intensity profile. The cluster intensity profile depicted in Figure 20 follows an attenuation pattern where the cluster signal is strongest at the cluster center and attenuates as one moves away from the cluster center. Subpopulations/groups/sets of clusters 1904 in a cluster population may have similar signal distributions. Clusters that share the same or similar signal distribution can be bucketed together (e.g. by grouping/addressing clusters according to their location coordinates) so that a professional signal profiler can be trained on that group/set of clusters representing that signal distribution. there is. Unlike spatial grouping, where spatially adjacent clusters are grouped, signal distribution-based grouping can group non-adjacent clusters. For example, edge clusters on opposite edges of a tile may have similar signal distributions and be grouped as having their intensity data calibrated by the same expert signal profiler (e.g., grouping clusters based on location coordinates). /by addressing).

As used herein, the phrase “similar signal distribution” refers to those signal distributions that share substantially overlapping signal patterns. For example, two signal patterns of similar shape (e.g., trapezoids) but different shape sizes (e.g., one larger trapezoid, one smaller trapezoid) may be considered to have similar signal distributions. Similarly, two signal patterns that have a common center or centers, for example within the range of 1 to 5 units in each dimension, may be considered to have similar signal distributions.

19, each/separate/different/independent expert signal profiler 1 to N 1908 is configured to analyze the signals of each signal distribution 1 to N 1902 corresponding to each cluster set 1 to N 1904. It is trained to maximize the noise-to-noise ratio. Of course, different signal distributions may be observed at different sequencing cycles, and thus different expert signal profilers may be trained and configured for application to different temporal stages of an ongoing sequencing run.

As used herein, the phrase “different temporal stages of an ongoing sequencing run” refers to different sequencing cycles or different groups of sequencing cycles of an ongoing sequencing run. For example, if a sequencing run has 150 sequencing cycles, each successive sequencing cycle may be at a different time stage, or sequencing cycle such as Cycles 1 to 20, Cycles 20 to 40, or Cycles 40 to 70. may be regarded as groups, etc., and thus as different time stages.

processing pipeline

Figure 21 illustrates one implementation of a processing pipeline implementing the disclosed technology. The processing pipeline may be implemented by real-time analysis module 225. The processing pipeline runs on a cycle-by-cycle basis 2100 according to one implementation, repeating 2102 for each new cycle. In one implementation, the input to the processing pipeline is tile images with a first (green) channel and a second (blue) channel.

At operation 2113, a template image is created that identifies the locations of clusters on the tile using sequencing images from some number of initial sequencing cycles, called template cycles. The template image is used as a reference for subsequent registration and intensity extraction steps. A template image is formed by detecting and merging bright spots in each sequencing image of a template cycle, which in turn sharpens the sequencing image 2114 (e.g. using a Laplacian convolution), a spatially separated Otsu approach. determining the “on” threshold by , and subsequent 5-pixel local maximum detection with subpixel position interpolation. The phrase “above threshold” may refer to an intensity value exceeding a preset value, for example 200 or 320, such that the intensity value is detected as being greater than the background intensity value or the noise intensity value.

The processing pipeline then matches the current sequencing image to the template image. This is achieved by using image correlation to align the current sequencing image to the template image of the subregion or by using a non-linear transformation (e.g., a full six-parameter linear affine transformation).

At operation 2115, the processing pipeline applies a non-linear distortion to each spot to account for optical distortions caused, for example, by the geometry of the optical lens. Non-linear distortion can be applied as channel-dependent coefficients of third-order polynomials.

At operation 2116, the processing pipeline segments the tile images into subtile images based on one or more configurations selected from different spatial configurations, temporal configurations, signal distribution configurations, or any combination thereof.

At operation 2118, the intensity is extracted from the segmented subtile images using corresponding expert signal profilers 1 to N.

In operation 2123, the subtile intensities are spatially normalized, for example by equalizing the 90th percentile of the extracted intensities.

At operation 2124, subtile intensities are compressed.

At operation 2125, the processing pipeline applies heuristic phase correction to compensate for noise in the image data caused by paging and pre-phasing errors.

At operation 2125, the processing pipeline spatially normalizes the extracted signal intensities to account for variations in illumination across the sampled imaging. For example, intensity values can be normalized such that percentiles 5 and 95 have values of 0 and 1, respectively. The normalized signal intensity for the image (e.g., the normalized intensity for each channel) can be used to calculate the average purity for multiple points in the image.

At operation 2133, the processing pipeline scales the intensities by cluster to account for variations in brightness of the clusters.

At operation 2134, the processing pipeline uses the Expectation Maximization (EM) algorithm to generate base calls as discussed above.

At operation 2135, the processing pipeline assigns quality scores to paid bases using quality tables (Q tables) 2152.

At operation 2136, the processing pipeline aligns the called bases to a reference genome (e.g., of a PhiX bacterium).

The processing pipeline includes base calling and quality scores 2128, InterOp files 2138 (binary reporting files for sequencing analysis views), and logs 2148 (e.g., error logs, general Generates certain outputs such as event log, processing event log, and warning event log.

different configuration

Other examples of configurations covered by the present invention include partitioning sequencing data and selecting library type, sample type, indexing type (first index read v/s second index read), read type (forward read v/s reverse read) ), read type (forward read v/s reverse read), physical properties of the sample, noise type (e.g., bubble), and reagent type.

Performance results - technical effectiveness and benefits as an objective indication of nonobviousness and inventive step

Figure 33 shows how the cost function for an expert signal profiler improves with each iteration of gradient descent. The cost function (or loss function) measures the performance of the model for given data. In Figure 33, different colored lines correspond to the number of subtiles into which the tile is divided, allowing the expert signal profiler to be adapted/trained/configured/updated independently for each subtile. As the number of subtiles increases from 1 to 16 (4x4 subtiles), the number of fitting parameters in the expert signal profiler also increases. As a result, the cost function is lower. The cost function is the sum of the squares of the Euclidean distance between well intensities and base call centroids for a well (cluster) sample. Improving this cost function also improves base calling accuracy.

Figure 34 is a plot showing the initial and final values of the cost function of Figure 33 as the expert signal profiler is adapted/trained/configured/updated at each sequencing cycle. In each successive sequencing cycle, we started with the same initial expert signal profiler and adapted the expert signal profiler using gradient descent. The plot shows that it is possible to adapt/train/configure/update a professional signal profiler at arbitrary sequencing cycles.

Figures 35A and 35B illustrate the improvement of primary analysis metrics for sequencing runs when adapting/training/configuring/updating an expert signal profiler. In this case, the average PhiX error rate improves from 0.3520% to 0.3316%.

Figures 36A and 36B show two plots evaluating the number of subtiles a sequencing tile can be divided into for adaptive equalization of each expert profiler. For each subtile, a separate expert signal profiler is adapted/trained/configured/updated using the wells from the corresponding subtile. Using more subtiles allows more accurate modeling of spatially varying phenomena within a tile, which is why the error rate and Q30 improve significantly as the number of subtiles from 1 to 9 increases. However, the number of wells available to adapt the model decreases as subtiles become smaller. In other words, this is why there are pros and cons to picking the right number of subtiles. In this specific case, increasing the number of subtiles from 9 to 16 slightly degrades the error rate and Q30.

computer system

37 depicts an example computer system 3700 that can be used to implement the disclosed techniques. Computer system 3700 includes at least one central processing unit (CPU) 3772 that communicates with a number of peripheral devices via a bus subsystem 3755. These peripheral devices include, for example, storage subsystem 3710, including memory devices and file storage subsystem 3736, user interface input devices 3738, user interface output devices 3776, and network interface. May include subsystem 3774. Input and output devices allow user interaction with computer system 3700. Network interface subsystem 3774 provides an interface to external networks, including an interface to corresponding interface devices in other computer systems.

In one implementation, executive signal profiler 3718 is communicatively linked to storage subsystem 3710 and user interface input devices 3738.

User interface input devices 3738 include a keyboard; Pointing devices such as a mouse, trackball, touchpad, or graphics tablet; scanner; Touch screen integrated within the display; audio input devices such as voice recognition systems and microphones; and other types of input devices. Broadly, use of the term “input device” is intended to include all possible types of devices and methods for inputting information into computer system 3700.

User interface output devices 3776 may include non-visual displays, such as a display subsystem, printer, fax machine, or audio output devices. The display subsystem may include a planar device such as an LED display, a cathode ray tube (CRT), a liquid crystal display (LCD), a projection device, or some other mechanism for producing a visible image. The display subsystem may also provide non-visual displays, such as audio output devices. Broadly, the use of the term “output device” is intended to include all possible types of devices and manners for outputting information from computer system 3700 to a user or to another machine or computer system.

Storage subsystem 3710 stores programming and data configurations that provide functionality of some or all of the modules and methods described herein. These software modules are generally executed by processors 3778.

Processors 3778 may be graphics processing units (GPUs), field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), and/or coarse-grained reconfigurable architectures (CGRAs). Processors 3778 may be hosted by deep learning cloud platforms such as Google Cloud Platform™, Xilinx™, and Cirrascale™. Examples of processors 3778 include Google's Tensor Processing Unit (TPU)™, rackmount solutions such as GX4 Rackmount Series™, GX37 Rackmount Series™, NVIDIA DGX-1™, Microsoft's Stratix V FPGA™, Graphcore's Intelligent Processor Unit (IPU)™, Qualcomm's Zeroth Platform™ with Snapdragon processors™, NVIDIA's Volta™, NVIDIA's DRIVE PX™, NVIDIA's JETSON TX1/TX2 MODULE™, Intel's Nirvana™, Movidius VPU™, Fujitsu DPI™ , ARM's DynamicIQ™, IBM TrueNorth™, and Lambda GPU servers with Testa VI 00s™.

The memory subsystem 3722 used in the storage subsystem 3710 includes a main random access memory (RAM) 3732 for storage of instructions and data during program execution and a read-only memory (ROM) where fixed instructions are stored. It may include multiple memories including (3734). File storage subsystem 3736 may provide persistent storage for program and data files and may include a hard disk drive, a floppy disk drive with associated removable media, a CD-ROM drive, an optical drive, or a removable media cartridge. You can. Modules implementing the functionality of certain implementations may be stored by file storage subsystem 3736 in storage subsystem 3710, or on other machines accessible by the processor.

Bus subsystem 3755 provides a mechanism for the various components and subsystems of computer system 3700 to communicate with each other as intended. Although bus subsystem 3755 is schematically depicted as a single bus, alternative implementations of the bus subsystem may use multiple buses.

Computer system 3700 itself may be a personal computer, portable computer, workstation, computer terminal, network computer, television, mainframe, server farm, widely distributed set of loosely networked computers, or any other data processing system or user device. It may be of various types including. Due to the ever-changing nature of computers and networks, the description of computer system 3700 depicted in FIG. 37 is intended only as a specific example to illustrate preferred implementations of the invention. Many other configurations of computer system 3700 are possible with more or fewer components than the computer system depicted in FIG. 37.

Neural network-based base caller

The following discussion focuses on the neural network-based base caller described herein that can be used with professional signal profilers. First, the input to a neural network-based base caller according to one implementation is described. Next, examples of the structure and form of neural network-based base callers are provided. Finally, the output of a neural network-based base caller according to one implementation is described.

Data flow logic provides sequencing images to a neural network-based base caller for base calling. A neural network-based base caller accesses sequencing images on a patch-by-patch (or tile-by-tile) basis. Each of the patches is a subgrid (or subarray) of pixelated units within a grid of pixelated units that form sequencing images. Patches have dimensions q x r of a subgrid of pixelated units, where q (width) and r (height) are arbitrary numbers ranging from 1 to 10,000 (e.g., 3 x 3, 5 x 5, 7 x 7, 10 x 10, 15 x 15, 25 x 25, 64 x 64, 78 x 78, 115 x 115). In some implementations, q and r are the same. In other implementations, q and r are different. In some implementations, patches extracted from a sequencing image are of the same size. In other implementations, the patches are of different sizes. In some implementations, patches may have overlapping pixelated units (eg, on edges).

Sequencing generates m sequencing images per sequencing cycle for the corresponding m image channels. That is, each of the sequencing images has one or more image (or intensity) channels (similar to the red, green, blue (RGB) channels of a color image). In one implementation, each image channel corresponds to one of a plurality of filter wavelength bands. In another implementation, each image channel corresponds to one of a plurality of imaging events in a sequencing cycle. In another implementation, each image channel corresponds to a combination of illumination using a specific laser and imaging through a specific optical filter. Image patches are tiled (or accessed) from each of the m image channels during a particular sequencing cycle. In different embodiments, such as 4-, 2-, and 1-channel chemistries, m is 4 or 2. In other embodiments, m is greater than 1, 3, or 4. In other implementations, the image may be in the blue and violet color channels instead of or in addition to the red and green channels.

For example, consider that a sequencing run is implemented using two different image channels: a blue channel and a green channel. Then, at each sequencing cycle, the sequencing run produces a blue image and a green image. In this way, during a series of k sequencing cycles of a sequencing run, sequences of k pairs of blue and green images are generated as output and stored as sequencing images. Therefore, for patch level processing by a neural network based base caller, a sequence of k pairs of blue and green image patches is generated.

The input image data for a neural network-based base caller for a single iteration of base calling (or a single forward traversal or a single instance of a forward pass) includes data for a sliding window of multiple sequencing cycles. The sliding window may include, for example, a current sequencing cycle, one or more preceding sequencing cycles, and one or more consecutive sequencing cycles.

In one implementation, the input image data includes data for three sequencing cycles, including data for the current (time t) sequencing cycle to be base called (i) left flanking/context/previous/previous/first. Accompanied by data for (time t -1) sequencing cycle, and (ii) right flanking/context/next/continuous/subsequent (time t +1) sequencing cycle.

In another implementation, the input image data includes data for 5 sequencing cycles, including data for the current (time t) sequencing cycle to be base called (i) first left flanking/context/previous/preceding/ Data for the first (time t -1) sequencing cycle, (ii) data for the second left flanking/context/previous/preceding/priority (time t -2) sequencing cycle, (iii) first right flanking/ Data for Context/Next/Continuous/Subsequent (Time t +1), and (iv) data for the second right flanking/Context/Next/Continuous/Subsequent (Time t +2) sequencing cycle.

In another embodiment, the input image data includes data for seven sequencing cycles, including data for the current (time t ) sequencing cycle to be base called, plus (i) the first left flank/context/prior /preceding/preceding (time t -1) data for the sequencing cycle, (ii) second left flanking/context/previous/preceding/preferring (time t -2) sequencing cycle, (iii) first Data for 3 left flanking/context/previous/previous/first (time t -3) sequencing cycles, (iv) data for the first right flanking/context/next/successive/successor (time t +1) , (iv) data for the second right flanking/context/next/continuous/subsequent (time t +2) sequencing cycle, and (v) data for the third right flanking/context/next/consecutive/subsequent (time t +3) Accompanied by data on sequencing cycles. In other implementations, the input image data includes data for a single sequencing cycle. In still other implementations, the input image data is 10, 15, 20, 30, 58, 75, 92, 130, 168, 175, 209, 225, 230, Includes data for 275, 318, 325, 330, 525, or 625 sequencing cycles.

A neural network-based base caller, according to one implementation, processes image patches through its convolutional layers and generates alternative representations. Then, an alternative expression is the immediately current (time t ) sequencing cycle or sequencing cycles, respectively, i.e. the current (time t ) sequencing cycle, the first and second preceding (time t -1, time t -2) sequencing cycles. , and an output layer (e.g., a softmax layer) to generate base calls during either the first and second trailing (time t +1, time t +2) sequencing cycles. The generated base calls form sequencing reads.

In one implementation, a neural network-based base caller outputs base calls for a single target cluster for a particular sequencing cycle. In another implementation, a neural network-based base caller outputs a base call for each target cluster within a plurality of target clusters during a particular sequencing cycle. In another embodiment, a neural network-based base caller outputs a base call for each target cluster within a plurality of target clusters during each sequencing cycle of the plurality of sequencing cycles, thereby generating a base call for each target cluster. Generate base calling sequence.

In one implementation, the neural network-based base caller is a multilayer perceptron (MLP). In another implementation, the neural network-based base caller is a feedforward neural network. In another implementation, the neural network based base caller is a fully connected neural network. In a further implementation, the neural network based base caller is a fully convolutional neural network. In another further implementation, the neural network based base caller is a semantic segmentation neural network. In yet a further implementation, the neural network-based base caller is a generative adversarial network (GAN).

In one implementation, the neural network-based base caller is a convolutional neural network (CNN) with multiple convolutional layers. In another implementation, the neural network-based base caller includes a short-term memory network (LSTM), a bi-directional LSTM (Bi-LSTM), or a gated recurrent unit (GRU). It is the same recurrent neural network (RNN). In another implementation, neural network based base callers include both CNNs and RNNs.

In still other implementations, the neural network based base caller can be used to perform an ID convolution, a 2D convolution, a 3D convolution, a 4D convolution, a 5D convolution, a dilated or atros convolution, a transpose convolution, a depth-wise separable convolution, Point-wise convolution, l x l convolution, group convolution, flat convolution, spatial and cross-channel convolution, shuffle grouped convolution, spatial separable convolution, and deconvolution can be used. A neural network-based base caller may use one or more loss functions, such as logistic regression/log loss, multi-class cross-entropy/softmax loss, and binary cross-entropy. cross-entropy loss, mean-squared error loss, L1 loss, L2 loss, smooth L1 loss, and Huber loss can be used. Neural network-based base callers support arbitrary parallelism, efficiency, and compression schemes, such as TFRecords, compressed encoding (e.g., PNG), sharding, parallel calls for map transformations, batching, and prefetching. (prefetching), model parallelism, data parallelism, and synchronous/asynchronous stochastic gradient descent (SGD) can be used. Neural network-based base callers include upsampling layers, downsampling layers, recurrent connections, gated and gated memory units (e.g., LSTM or GRU), residual blocks, residual connections, highway connections, skip connections, peephole connections, and activation. Functions (e.g., nonlinear transformation functions such as rectifying linear unit (ReLU), leaky ReLU, exponential liner unit (ELU), sigmoid, and hyperbolic tangent (tanh)), batch normalization layer, regularization layer, drop It may include an out, pooling layer (e.g., maximum or average pooling), a global average pooling layer, and an attenuation mechanism.

A neural network-based base caller is trained using backpropagation-based gradient update techniques. Exemplary gradient descent techniques that can be used to train a neural network-based base caller include stochastic gradient descent, batch gradient descent, and mini-batch gradient descent. Some examples of gradient descent optimization algorithms that can be used to train a neural network-based base caller are Momentum, Nesterov accelerated gradient, Adagrad, Adadelta, RMSprop, Adam, AdaMax, Nadam, and AMSGrad.

In one implementation, a neural network-based base caller uses a specialized architecture to segregate the processing of data during different sequencing cycles. The motivation for using a specialized architecture is first described. As discussed above, the neural network-based base caller processes image patches during the current sequencing cycle, one or more preceding sequencing cycles, and one or more successive sequencing cycles. Data on additional sequencing cycles provide sequence-specific context. Neural network-based base callers learn sequence-specific contexts during training and base call them. Moreover, data on pre- and post-sequencing cycles provide secondary contributions of pre-phasing and phasing signals to the current sequencing cycle.

However, images captured in different sequencing cycles and in different image channels are misaligned and have residual registration errors with respect to each other. To handle such misalignments, specialized architectures include spatial convolution layers that do not mix information between sequencing cycles, but only mix information within one sequencing cycle.

Spatial convolution layers (or spatial logic) use so-called "separate convolutions" that allow separation to be manipulated by independently processing the data during each of multiple sequencing cycles through a "dedicated, non-shared" sequence of convolutions. use. Separate convolutions convolution the data and generated feature maps only for a given sequencing cycle, i.e., intra-cycle, without convolving over the data and generated feature maps of any other sequencing cycle. Routine.

For example, the input image data can be (i) the current image patch during the current (time t ) sequencing cycle to be base called, (ii) the previous image patch during the previous (time t -1) sequencing cycle, and (iii) ) is considered to include the next image patch for the next (time t +1) sequencing cycle. The specialized architecture then starts three separate convolution pipelines: the current convolution pipeline, the previous convolution pipeline, and the next convolution pipeline. The current data processing pipeline receives the current image patch as input during the current (time t ) sequencing cycle and independently processes it through a plurality of spatial convolution layers to produce the so-called “current spatial convolution layer” as the output of the final spatial convolution layer. Creates a “convolved representation”. The previous convolution pipeline receives as input the previous image patch during the previous (time t -1) sequencing cycle, processes it independently through a plurality of spatial convolution layers, and produces the so-called "previous" patch as the output of the final spatial convolution layer. Creates a “spatially convolved representation”. The next convolution pipeline receives the next image patch as input during the next (time t +1) sequencing cycle and independently processes it through a plurality of spatial convolution layers to produce the so-called “next” as the output of the final spatial convolution layer. Creates a “spatially convolved representation”.

In some implementations, the current, previous, and next convolution pipelines run in parallel. In some implementations, spatial convolutional layers are part of a spatial convolutional network (or subnetwork) within a specialized architecture.

The neural network-based base caller further includes temporal convolution layers (or temporal logic) that mix information between sequencing cycles, i.e., inter-cycle. The temporal convolutional layers receive their inputs from the spatial convolutional network and operate on the spatially convolutional representations produced by the final spatial convolutional layer for their respective data processing pipelines.

The cycle-to-cycle operability freedom of the temporal convolutional layers is such that the misalignment properties present in the image data supplied as input to the spatial convolutional network are achieved by a stack or cascade of disjoint convolutions performed by a sequence of spatial convolutional layers. It follows from the fact that it is purged out from spatially convolved representations.

The temporal convolution layer uses so-called “combinatorial convolutions”, which group-by-group convolve over input channels in successive inputs on a sliding window basis. In one implementation, the continuous inputs are continuous outputs produced by the previous spatial convolution layer or the previous temporal convolution layer.

In some implementations, temporal convolutional layers are part of a temporal convolutional network (or subnetwork) within a specialized architecture. A temporal convolutional network receives its inputs from a spatial convolutional network. In one implementation, the first temporal convolutional layer of the temporal convolutional network combines spatially convolved representations group by group between sequencing cycles. In another implementation, subsequent temporal convolutional layers of a temporal convolutional network combine successive outputs of previous temporal convolutional layers. The output of the final temporal convolution layer is fed to the output layer that generates the output. The output is used to base call one or more clusters in one or more sequencing cycles.

Additional details regarding the neural network-based base caller can be found in U.S. Provisional Patent Application Serial No. 62/821,766, entitled “Artificial Intelligence-Based Sequencing,” filed March 21, 2019 (Attorney Docket No. ILLM 1008-9/IP) - 1752-PRV), which is incorporated herein by reference.

item

The following items are part of this disclosure:

One. As a system,

A memory storing a plurality of specialized signal profilers, wherein each specialized signal profiler in the plurality of specialized signal profilers is a specific signal profile detected for an analyte in a specific analyte class and characterized in a specific training data set. a memory trained to maximize the signal-to-noise ratio of the sequenced signal;

Runtime logic that accesses the memory, wherein during a base calling operation, a sequenced signal from each detected signal profile for an analyte within each analyte class is accessed by each expert within the plurality of expert signal profilers. A system comprising runtime logic configured to execute base calling operations by applying a signal profiler.

2. The system of item 1, wherein each analyte class represents a different spatial configuration of the analytes that contribute to the generation of the respective signal profile during the base calling operation.

3. The system of item 2, wherein the different spatial configurations include analytes located on different surfaces of the biosensor on which the base calling operation is performed.

4. The system of item 3, wherein the different surfaces include an upper surface and a lower surface.

5. The system of any of items 2-4, wherein the different spatial configurations include analytes located on different lanes of the biosensor.

6. The system of any of items 2-5, wherein the different spatial configurations include analytes located on different lane groups of the biosensor.

7. The system of item 6, wherein the different lane groups include upper perimeter lanes, central lanes, and lower perimeter lanes.

8. The system of item 6 or item 7, wherein the different lane groups include edge lanes and non-edge lanes.

9. The system of any of items 2-8, wherein the different spatial configurations include analytes located on different swaths of the different lanes of the biosensor.

10. The system of item 9, wherein the different swaths include a top perimeter swath, a center swath, and a bottom perimeter swath.

11. The system of item 9 or item 10, wherein the different swaths include an edge swath and a center swath.

12. The system of any of items 9-11, wherein the different spatial configurations include analytes located on different tiles of the different swaths of the different lanes of the biosensor.

13. The system of any of items 2-12, wherein the different spatial configurations include analytes located on different tile groups of the biosensor.

14. The system of item 13, wherein the different tile groups include edge tiles, center tiles, and near-edge tiles.

15. The system of any of items 12-14, wherein the different spatial configurations include analytes located on different subtiles of the different tiles of the different swaths of the different lanes of the biosensor.

16. The system of any of items 2-15, wherein the different spatial configurations include analytes located on different sections of the biosensor.

17. The system of item 16, wherein the different sections include an upper right section, an upper center section, an upper left section, a middle right section, a center section, a middle left section, a lower left section, a lower center section, and a lower left section.

18. The method of any one of items 1-17, wherein each expert signal profiler performs a signal-to-signal analysis of signals detected for an analyte within a specific analyte subclass and sequenced in a specific signal profile characterized in a specific training data subset. are further trained to maximize noise ratio,

The runtime logic is further configured to execute the base calling operation by applying the respective specialized signal profiler to the sequenced signal in each signal profile detected for the analyte within each analyte subclass during the base calling operation. Being a system.

19. The method of item 18, wherein each analyte subclass represents a different spatial configuration of the analyte that produced the sequenced signal at a different time period of the base calling operation, and a different combination of the different spatial configuration and the different time period. contributes to the generation of each detected signal profile during the base calling operation.

20. The system of item 19, wherein the different time periods correspond to different sensing cycles in the series of sensing cycles of the base calling operation.

21. The system of item 19 or item 20, wherein the different time periods correspond to different subseries of sense cycles in the series of sense cycles of the base call operation.

22. The system of any of items 1 to 21, wherein each expert signal profiler is comprised of channel-specific equalizers, and each channel-specific equalizer has a plurality of convolution kernels.

23. The system of any of items 1 to 22, wherein the runtime logic is further configured to iteratively train each expert signal profiler during the base call operation.

24. Item 23, wherein, for the current training iteration, the runtime logic sets an expectation that iteratively maximizes the probability of observing, by channel, the signal distribution and base-by-base signal centroid that best fits the sequenced signal detected so far during the base calling operation. Implement maximization, determine on a channel-by-channel basis a signal-to-noise ratio-maximizing sequenced signal in response to applying the respective expert signal profiler to the sequenced signal, and call bases based on the signal-to-noise ratio-maximizing sequenced signal. and determining a base calling error for each channel by comparing the signal-to-noise-ratio-maximizing sequenced signal with the signal center of the called base,

The system further configured to update convolution kernel coefficients of each expert signal profiler on a channel-by-channel basis based on the base call error.

25. The system of any of items 1 to 24, wherein the analytes correspond to wells when the biosensor is a patterned biosensor.

26. The system of any one of items 1 through 25, wherein the sequenced signal is an intensity signal.

27. The system of any of items 1 through 25, wherein the sequenced signal is a voltage signal.

28. The system of any one of items 1 through 25, wherein the sequenced signal is an intensity signal.

29. As a system,

a memory storing initially sequenced signals detected during initial sequencing cycles of a sequencing run;

fitting logic accessing the memory, the fitting logic configured to fit a plurality of signal distributions on the initially sequenced signals and store the plurality of signal distributions in the memory;

Online training logic accessing the memory, training each expert signal profiler in the plurality of expert signal profilers to maximize the signal-to-noise ratio of each signal distribution in the plurality of signal distributions and storing each trained signal profiler in the memory. Online training logic configured to store expert signal profilers; and

Runtime logic that accesses the memory to uniquely map a subsequently sequenced signal detected during a subsequent sequencing cycle of the sequence run to the respective signal distribution, and to the unique mapping for the respective signal distribution. and runtime logic configured to apply each trained expert signal profiler to the subsequently sequenced signal to generate base calls for the subsequent sequencing cycle.

30. The system of item 29, wherein at least some of the signal distributions within the plurality of signal distributions represent different underlying sequencing events contributing to the generation of the some of the signal distributions.

31. The system of item 30, wherein the basic sequencing events include the formation of bubbles on the biosensor on which the sequencing run is run.

32. The system of any of items 29-31, wherein at least some of the signal distributions in the plurality of signal distributions represent different analyte locations on the biosensor that contribute to the generation of the some of the signal distributions. .

33. The system of any of items 29-32, wherein at least some of the signal distributions in the plurality of signal distributions represent different sequencing cycles of the sequencing run contributing to the generation of the some of the signal distributions.

34. The system of any of items 29-33, wherein each of the trained expert signal profilers has respective sets of coefficients of variation corresponding to the respective signal distribution.

35. The system of item 34, wherein each set of variation coefficients comprises channel-specific amplification coefficients that correct for scale variations in the corresponding signal distribution.

36. The system of item 34 or item 35, wherein each set of variation coefficients comprises channel-specific offset coefficients that correct shift variations in the corresponding signal distribution.

37. The method of any one of items 29 to 36, further configured to include control logic configured to repeat execution of the fitting logic, the online training logic, and the runtime logic in each sequencing cycle of the sequencing run. system.

38. The method of any one of items 29 to 37, further comprising control logic configured to repeat execution of the fitting logic, the online training logic, and the runtime logic after a predetermined number of sequencing cycles of the sequencing run. Consisting of a system.

39. The method of any one of items 29 to 38, further comprising control logic configured to repeat execution of the fitting logic, the online training logic, and the runtime logic at predetermined sequencing cycles of the sequencing run. Consisting of a system.

40. As a system,

A memory configured to store a plurality of expert signal profilers configured for use in a base calling operation, wherein each expert signal profiler in the plurality of expert signal profilers is observed at a respective sequencing event of the base calling operation and generates respective training data. A memory is trained to maximize the signal-to-noise ratio of the sensor data for each signal distribution characterized in the set; and

Runtime logic that accesses the memory to select an expert signal profiler from the plurality of expert signal profilers based on a subject sequencing event that generated subject sensor data, and apply the selected expert signal profiler on the subject sensor data. A system, comprising runtime logic configured to apply and generate base call classification data for target sequencing events.

41. The method of item 40, wherein the signal-to-noise ratio of the sensor data generated by each sequencing event degrades with the temporal progression of the base calling operation, and is selected based on the respective sequencing event and the temporal progression of the base calling operation. Each expert signal profiler applied over time reverses the degradation of the signal-to-noise ratio of the sensor data.

42. The system of item 41, wherein each sequencing event is a temporal progression of the base calling operation through each sensing cycle in a series of sensing cycles of the base calling operation.

43. The system of item 42, wherein each sequencing event is a temporal progression of the base calling operation through each subseries of sensing cycles in the series of sensing cycles.

44. The system of item 41, wherein each sequencing event is a spatial progression of the base calling operation through respective analyte positions on the biosensor on which the base calling operation is performed.

45. The system of item 40, wherein the runtime logic is further configured to iteratively train each expert signal profiler during the sequencing run at each sequencing event.

46. In item 45, for the current training iteration, the runtime logic implements expectation maximization to iteratively maximize the probability of observing the signal distribution and base-by-base signal center that best fits the sensor data, channel by channel, and Responsive to applying a signal profiler to the sensor data, determine signal-to-noise ratio maximizing sensor data on a channel-by-channel basis, call bases based on the signal-to-noise ratio maximization sensor data, and call bases based on the signal-to-noise ratio maximization sensor data. The system further configured to determine a base call error on a channel-by-channel basis by comparing it to the signal center of the base, and update the convolution kernel coefficients of each expert signal profiler on a channel-by-channel basis based on the base call error.

47. As a system,

a memory storing initially sequenced signals detected during initial sequencing cycles of a sequencing run for an analyte population;

Fitting logic accessing the memory, processing the initially sequenced signal for each analyte, fitting each signal profile for each analyte in the analyte population, and storing each signal profile in the memory. fitting logic configured to store;

Online training logic accessing the memory, training each expert signal profiler in a plurality of expert signal profilers to maximize the signal-to-noise ratio of each fitted signal profile for each analyte and storing the signal profile in the memory. online training logic configured to store each trained expert signal profiler; and

Runtime logic that accesses the memory to uniquely map subsequently sequenced signals detected during a subsequent sequencing cycle of the sequence run by analyte to each signal profile, The system comprising runtime logic configured to apply each trained expert signal profiler to the subsequently sequenced signal based on the mapping to generate analyte-specific base calls for the subsequent sequencing cycle.

48. The system of item 47, wherein each of the trained expert signal profilers has respective sets of variation coefficients corresponding to the respective signal profile.

49. The system of item 48, wherein each set of coefficients of variation includes channel-specific amplification coefficients that correct scale variations in the corresponding signal profile.

50. The system of item 48, wherein each set of variation coefficients includes channel-specific offset coefficients that correct for shift variations in the corresponding signal profile.

51. The system of item 47, further configured to include control logic configured to repeat execution of the fitting logic, the online training logic, and the runtime logic in each sequencing cycle of the sequencing run.

52. The system of item 51, further configured to include control logic configured to repeat execution of the fitting logic, the online training logic, and the runtime logic after a predetermined number of sequencing cycles of the sequencing run.

53. The system of item 51, further configured to include control logic configured to repeat execution of the fitting logic, the online training logic, and the runtime logic at predetermined sequencing cycles of the sequencing run.

54. As a system,

A memory storing a plurality of expert signal profilers, wherein each expert signal profiler in the plurality of expert signal profilers represents a signal sequenced in a particular signal profile detected for a particular analyte and characterized in a particular training data set. Memory trained to maximize signal-to-noise ratio; and

Runtime logic to access the memory, wherein the base call is performed by applying each expert signal profiler in the plurality of expert signal profilers to a sequenced signal in each signal profile detected for each analyte during a base call operation. A system, including runtime logic configured to execute operations.

55. As a system,

Spatial classification logic configured to segment an analyte population into spatial classes using analyte location, each spatial class comprising a non-overlapping set of analytes culled from the analyte population;

Signal profiling logic configured to estimate one or more signal profiles for each spatial class using the sequencing signals detected for the population of analytes, wherein each signal profile is one of analytes culled from a non-overlapping set of analytes. signal profiling logic including non-overlapping subsets; online training logic configured to train at least one expert signal profiler to maximize the signal-to-noise ratio of each signal profile estimated for each spatial class; and

A system comprising runtime logic configured to base call analytes within respective non-overlapping subsets of analytes using respective trained expert signal profilers.

56. The system of item 55, wherein the online training logic is further configured to match training data used to train a particular expert signal profiler to detected sequencing signals for a particular non-overlapping subset of analytes.

57. The system of item 55 or item 56, wherein certain non-overlapping subsets of analytes are further configured to adjacently merge adjacent non-overlapping subsets of analytes when no suitable analytes are available to make appropriate training data available.

58. The system of any of items 55-57, wherein the number of analytes within each non-overlapping subset of analytes is configurable to optimize training.

59. The system of any of items 55-58, wherein the online training logic is further configured to estimate a set of coefficients of variation for each trained expert signal profiler.

60. The system of item 59, wherein the set of variation coefficients comprise channel-specific amplification coefficients that correct scale variations in the corresponding signal profile.

61. The system of item 59 or item 60, wherein the set of variation coefficients comprise channel-specific offset coefficients that correct shift variations in the corresponding signal profile.

62. The method of any one of items 55 to 61, further configured to include control logic configured to repeat execution of the signal profiling logic, the online training logic, and the runtime logic at each sequencing cycle of a sequencing run. , system.

63. The method of any one of items 55-62, comprising control logic configured to repeat execution of the signal profiling logic, the online training logic, and the runtime logic after a predetermined number of sequencing cycles of the sequencing run. A system further configured to:

64. The method of any one of items 55 to 63, comprising control logic configured to repeat execution of the signal profiling logic, the online training logic, and the runtime logic at predetermined sequencing cycles of the sequencing run. A system further configured to:

Claims (34)

As a system,
A memory storing a plurality of specialized signal profilers, wherein each specialized signal profiler in the plurality of specialized signal profilers is a specific signal profile detected for an analyte in a specific analyte class and characterized in a specific training data set. a memory trained to maximize the signal-to-noise ratio of the sequenced signal;
Runtime logic that accesses the memory, wherein during a base calling operation, a sequenced signal from each detected signal profile for an analyte within each analyte class is accessed by each expert within the plurality of expert signal profilers. Runtime logic configured to execute base calling operations by applying a signal profiler.
system, including. The system of claim 1, wherein each analyte class represents a different spatial configuration of the analytes that contribute to the generation of the respective signal profile during the base calling operation. 3. The system of claim 2, wherein the different spatial configurations include analytes located on different surfaces of the biosensor on which the base calling operation is performed. 4. The system of claim 3, wherein the different surfaces include an upper surface and a lower surface. 5. The system of any one of claims 2-4, wherein the different spatial configurations include analytes located on different lanes of the biosensor. 6. The system of any one of claims 2-5, wherein the different spatial configurations comprise analytes located on different lane groups of the biosensor. 7. The system of claim 6, wherein the different lane groups include an upper perimeter lane, a central lane, and a lower perimeter lane. 8. The system of claim 6 or 7, wherein the different lane groups include edge lanes and non-edge lanes. 9. The system of any one of claims 2-8, wherein the different spatial configurations include analytes located on different swaths of the different lanes of the biosensor. 10. The system of claim 9, wherein the different swaths include a top perimeter swath, a center swath, and a bottom perimeter swath. 11. The system of claim 9 or 10, wherein the different swaths include an edge swath and a center swath. 12. The system of any one of claims 9-11, wherein the different spatial configurations include analytes located on different tiles of the different swaths of the different lanes of the biosensor. 13. The system of any one of claims 2-12, wherein the different spatial configurations include analytes located on different tile groups of the biosensor. 14. The system of claim 13, wherein the different tile groups include edge tiles, center tiles, and near-edge tiles. 15. The system of any one of claims 12-14, wherein the different spatial configurations include analytes located on different subtiles of the different tiles of the different swaths of the different lanes of the biosensor. . 16. The system of any one of claims 2-15, wherein the different spatial configurations include analytes located on different sections of the biosensor. 17. The system of claim 16, wherein the different sections include an upper right section, an upper center section, an upper left section, a middle right section, a center section, a middle left section, a lower left section, a lower center section, and a lower left section. . 18. The method of any one of claims 1 to 17, wherein each expert signal profiler is configured to: further trained to maximize signal-to-noise ratio,
The runtime logic is further configured to execute the base calling operation by applying the respective specialized signal profiler to the sequenced signal in each signal profile detected for the analyte within each analyte subclass during the base calling operation. Being a system. 19. The method of claim 18, wherein each analyte subclass represents a different spatial configuration of the analyte that produced the sequenced signal at a different time period of the base calling operation, and wherein the different spatial configuration and the different time period are A system wherein different combinations contribute to the generation of each detected signal profile during the base calling operation. 20. The system of claim 19, wherein the different time periods correspond to different sensing cycles in the series of sensing cycles of the base calling operation. 21. The system of claim 19 or 20, wherein the different time periods correspond to different subseries of sensing cycles in the series of sensing cycles of the base calling operation. 22. The system according to any one of claims 1 to 21, wherein each specialized signal profiler consists of a channel-specific equalizer, and each channel-specific equalizer has a plurality of convolution kernels. 23. The system of any preceding claim, wherein the runtime logic is further configured to iteratively train each expert signal profiler during the base call operation. 24. The method of claim 23, wherein, for a current training iteration, the runtime logic iteratively maximizes the likelihood of observing, by channel, the signal distribution and base-by-base signal centroid that best fits the sequenced signal detected so far during the base calling operation. Implement expectation maximization, determine on a channel-by-channel basis a signal-to-noise ratio-maximizing sequenced signal in response to applying the respective expert signal profiler to the sequenced signal, and select bases based on the signal-to-noise ratio-maximizing sequenced signal. calling, and comparing the signal-to-noise-ratio-maximizing sequenced signal to the signal center of the called base to determine a base calling error for each channel;
The system further configured to update convolution kernel coefficients of each expert signal profiler on a channel-by-channel basis based on the base call error. 25. The system of any one of claims 1-24, wherein the analytes correspond to wells when the biosensor is a patterned biosensor. 26. The system of any preceding claim, wherein the sequenced signal is an intensity signal. 26. The system of any preceding claim, wherein the sequenced signal is a voltage signal. 26. The system of any preceding claim, wherein the sequenced signal is a current signal. As a system,
a memory storing initially sequenced signals detected during initial sequencing cycles of a sequencing run;
fitting logic accessing the memory, the fitting logic configured to fit a plurality of signal distributions on the initially sequenced signals and store the plurality of signal distributions in the memory;
Online training logic accessing the memory, training each expert signal profiler in the plurality of expert signal profilers to maximize the signal-to-noise ratio of each signal distribution in the plurality of signal distributions and storing each trained signal profiler in the memory. Online training logic configured to store an expert signal profiler of; and
Runtime logic that accesses the memory to uniquely map a subsequently sequenced signal detected during a subsequent sequencing cycle of the sequence run to the respective signal distribution, and to the unique mapping for the respective signal distribution. and runtime logic configured to apply each trained expert signal profiler to the subsequently sequenced signal to generate base calls for the subsequent sequencing cycle. 30. The system of claim 29, wherein at least some of the signal distributions within the plurality of signal distributions represent different underlying sequencing events contributing to the generation of the some of the signal distributions. As a system,
A memory configured to store a plurality of expert signal profilers configured for use in a base calling operation, wherein each expert signal profiler in the plurality of expert signal profilers is observed and trained at a respective sequencing event of the base calling operation. a memory trained to maximize the signal-to-noise ratio of the sensor data for each signal distribution characterized in the data set; and
Runtime logic that accesses the memory to select an expert signal profiler from the plurality of expert signal profilers based on a subject sequencing event that generated subject sensor data, and apply the selected expert signal profiler on the subject sensor data. A system comprising runtime logic configured to apply and generate base call classification data for the target sequencing event. As a system,
a memory storing initially sequenced signals detected during initial sequencing cycles of a sequencing run for an analyte population;
Fitting logic accessing the memory, processing the initially sequenced signal for each analyte, fitting each signal profile for each analyte in the analyte population, and storing each signal profile in the memory. fitting logic configured to store;
Online training logic accessing the memory, training each expert signal profiler in a plurality of expert signal profilers to maximize the signal-to-noise ratio of each fitted signal profile for each analyte and storing the signal profile in the memory. online training logic configured to store each trained expert signal profiler; and
Runtime logic that accesses the memory to uniquely map subsequently sequenced signals detected during a subsequent sequencing cycle of the sequence run for each analyte to each signal profile, and A system comprising runtime logic configured to apply each trained expert signal profiler to the subsequently sequenced signal based on a unique mapping to generate base calls for each of the analytes for the subsequent sequencing cycle. . As a system,
A memory storing a plurality of expert signal profilers, wherein each expert signal profiler in the plurality of expert signal profilers represents a signal sequenced in a particular signal profile detected for a particular analyte and characterized in a particular training data set. Memory trained to maximize signal-to-noise ratio; and
Runtime logic that accesses the memory to perform base calling by applying each specialized signal profiler in the plurality of specialized signal profilers to the sequenced signal in each signal profile detected for each analyte during the base calling operation. A system, including runtime logic configured to execute operations. As a system,
Spatial classification logic configured to segment an analyte population into spatial classes using analyte location, each spatial class comprising a non-overlapping set of analytes culled from the analyte population;
Signal profiling logic configured to estimate one or more signal profiles for each spatial class using the sequencing signals detected for the population of analytes, wherein each signal profile is culled from a non-overlapping set of analytes. signal profiling logic comprising a non-overlapping subset of; online training logic configured to train at least one expert signal profiler to maximize the signal-to-noise ratio of each signal profile estimated for each spatial class; and
A system comprising runtime logic configured to base call analytes within respective non-overlapping subsets of analytes using respective trained expert signal profilers.

Images