CN106295251A - Phenotypic data analysis and processing method based on unicellular Phenotype data base - Google Patents

Abstract

The invention relates to a method for analyzing and processing phenotype data based on a single-cell phenotype database. The main module of the present invention consists of two parts: single cell phenotype database and phenotype data analysis and processing method. (1) A cell image analysis and processing method based on a single-cell phenotype database. This method analyzes and compares the cell image information in the phenotype database, compares the unknown cell image data, and finds the best matching cell to obtain the unknown cell. Cell details. (2) Analysis and processing method of cell Raman data based on single cell phenotype database. The method obtains the detailed information of the unknown cell by analyzing and comparing the cell Raman information in the phenotype database, combining the Raman data of the unknown cell for comparison and finding the best matching cell.

Description

Phenotype data analysis and processing method based on single-cell phenotype database

technical field

The invention relates to the fields of single cell research and cell science application, in particular to a method for analyzing and processing phenotype data based on a single cell phenotype database.

Background technique

A single cell is the basic unit of life activities. All organisms on earth are composed of single cells or differentiated from single cells. In-depth and systematic research on single cells can not only reveal the essence of life activities in a panoramic manner, but also the specificity and differentiation process of single cells is of great significance for the study of disease mechanisms and diagnosis and prevention of diseases. "Single-cell research" (analysis of single cells for specific functions) will be able to analyze the "deepest" operating mechanism of living systems, thus bringing life science and its wide application in energy, environment, health, agriculture, oceans, etc. field breakthrough. The National Institutes of Health (NIH) launched the "Single Cell Analysis Program" in September 2012, announcing a total of 90 million US dollars in funding for 26 projects, mainly for the development of new tools and technologies in the field of single cells (http://commonfund.nih.gov/singlecell/fundedresearch.aspx.). On December 21, 2012, Science magazine selected single-cell research as one of the six most noteworthy scientific fields in 2013.

The phenotype of the cell is also the form of expression of the cell, that is, the information that can be obtained by using the overall observation method and reflects the growth state of the cell. For a single cell, something that expresses its particular physical appearance or composition, such as cell shape, size, color characteristics, texture characteristics, class, etc., are examples of phenotypes. The important methods include microscopic identification of single cell morphology, and cell Raman spectrum signals obtained by Raman spectrometer and other equipment. The research on single cells, that is, the analysis of cell shape, size, color and other related information and the discrimination of cell types, all of which need to rely on a phenotype database containing different cells and different growth cycles and corresponding phenotype data analysis and processing system to achieve. At present, there are few related studies on phenotype data analysis and processing systems in China, which makes the establishment of a set of phenotype data analysis and processing methods based on single-cell phenotype databases have important practical value for the study of single cells.

Contents of the invention

Aiming at the above-mentioned deficiencies in the prior art, the technical problem to be solved in the present invention is to provide a method for analyzing and processing phenotype data based on a single-cell phenotype database, and obtain a single or group cell through a new generation of cell sorting equipment. Phenotype information of cells (applicable to microorganisms, plants, animals or human cells), thus laying a fundamental foundation for the omics analysis, transformation and utilization of these cells.

The technical solution adopted by the present invention to achieve the above object is: a method for analyzing and processing phenotype data based on a single-cell phenotype database, comprising the following steps:

Cell image analysis and processing stage: by analyzing and comparing cell image information and unknown cell image data in the phenotype database; extracting phenotypic characteristics of unknown cells;

Data preprocessing: process the extracted phenotypic features into data suitable for processing by Euclidean distance algorithm, KNN algorithm, and support vector machine algorithm;

Classification analysis is performed based on the characteristic data in the single-cell phenotype database to find the best matching cells.

The data preprocessing includes the following steps:

Perform grayscale transformation on the image;

Perform image sharpening to enhance the grayscale contrast, thereby enhancing the edge information in the image;

Smoothing and filtering the image to remove noise sources;

Find the place with the largest grayscale change rate in the image, get the closed contour of the cell image, and then extract the features in the contour.

The image is smoothed and filtered using digital Fourier filtering, specifically:

First, fast Fourier transform is performed on the data, and the Gaussian window function is applied to the frequency space, and then the fast Fourier transform is reversed to obtain spectral data after bandpass filtering.

The classification analysis based on the characteristic data in the single-cell phenotype database includes two stages of training and judgment;

First, obtain typical samples of known cell phenotypes, perform feature extraction and data preprocessing, obtain feature samples to train the SVM model, find support vectors in the training samples, and determine the parameters of the SVM model; then, input unknown cell information into The already parameterized SVM model judges it.

The present invention has the following advantages and beneficial effects: collecting different types of single-cell samples, by constructing a single-cell phenotype database system, with the help of phenotype data analysis and processing means, to discriminate unknown cell types and phenotypic characteristics, overcoming the inability to identify unknown The bottleneck of cell type discrimination, and with the help of a new generation of cell sorting equipment, in situ and real-time cell type discrimination can be realized, which is easy to popularize in the market. The application of the present invention will speed up the research of single cell analysis.

Description of drawings

Figure 1. Cell sample feature extraction; cell sample feature extraction is performed on the cell image by grayscale transformation, boundary detection, depth-first search, etc., to obtain the position of each cell, and then perform feature extraction and data analysis on each cell according to the position. Save, so as to facilitate subsequent operations such as classification and identification.

Figure 2. The recognition process of the support vector machine; SVM trains the SVM model according to the phenotype characteristics of the known cells to determine the parameters of the SVM model, and then analyzes and processes the characteristic information of the unknown cells based on the model parameters to determine the detailed information of the unknown cells .

detailed description

The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments.

The present invention develops a method for analyzing and processing phenotypic data based on a single-cell phenotype database, collects different types of single-cell samples, constructs a single-cell phenotype database system, and uses phenotype data analysis and processing means to analyze unknown cell types and Identify phenotypic characteristics. It mainly includes the following two aspects (microalgae samples are used as examples below):

1. A cell image analysis and processing method based on a single-cell phenotype database. The method obtains the detailed information of the unknown cell by analyzing and comparing the cell image information in the phenotype database, combining the image data of the unknown cell for comparison and finding the best matching cell. The specific process is as follows:

(1) Training sample feature extraction and sample set construction

Extracting cell phenotype characteristics is a key link in cell image analysis and processing, and it is also the basis for constructing sample sets. Once the raw data of cell phenotype are obtained, the raw data should be mapped to points or vectors in the sample space. These data include the physiological intrinsic appearance information of each observation part of the cell phenotype, such as changes and intensity of vision, touch, etc. The combination of appearance and intensity is the key basis for determining unknown cell types.

Generally speaking, raw data contains redundant information, which needs to be processed and transformed in order to effectively extract cell phenotype characteristics. The process of processing the extracted sample data of cell phenotype characteristics into data suitable for support vector machine processing is called data preprocessing. First, the image is gray-scale transformed; then the image is sharpened to enhance the gray-scale contrast, thereby enhancing the edge information in the image, which is conducive to contour extraction; and then smoothing the image to filter out noise sources (such as electronic noise). , photon noise, speckle noise and quantization noise, etc.), so as to improve the signal-to-noise ratio of the image and facilitate the search of the image contour; finally, find the place with the largest gray scale change rate in the image, so as to obtain the closed contour of the cell image, and then extract the contour features in (Figure 1). The corresponding data can then be stored in a single-cell phenotype database based on the characteristic values.

(2) Recognition process of support vector machine

Classification analysis is then performed based on the feature values in the database. The classification algorithms currently developed and applied include Euclidean distance algorithm, KNN algorithm, and support vector machine (SVM) algorithm. Taking the SVM algorithm as an example, the implementation is carried out in two stages: training and judgment. First, obtain typical samples of known cell phenotypes, perform feature extraction and data preprocessing, obtain feature samples to train the SVM model, find support vectors in the training samples, and determine the parameters of the SVM model; then, input unknown cell information into The parameterized SVM model judges it, and the specific flow chart is shown in Figure 2.

2. Cellular Raman data analysis and processing method based on single-cell phenotype database. The method obtains the detailed information of the unknown cell by analyzing and comparing the cell Raman information in the phenotype database, combining the Raman data of the unknown cell for comparison and finding the best matching cell. The specific process is as follows:

2.1 Spectral processing module

Starting from the structure and working principle of the Raman system, the main factors affecting the spectral signal mainly include the following aspects:

(1) Interference signals from the optical system and detection objects

The interference signal of the optical system to the spectral signal is mainly the false signal caused by the system's stray light, aberration and useless secondary spectral level. For traditional large-scale spectrometers, the interference signal of the optical system is mainly eliminated from the system structure and related optical components. However, the structure, optical components and system integration of the laser Raman spectrometer are greatly restricted, and the traditional method is not feasible. Only by studying the basic characteristics of these interference signals, and then seek solutions.

(2) Noise signals from circuit system and power supply

The drift and fluctuating signal of the signal acquisition and processing circuit, and the power supply noise signal are also the main sources of the interference signal. In particular, the above interference signals have a great influence when the signal is weak, and sometimes may completely submerge useful signals, seriously affecting the detection performance of the system. The processing of this part of the signal is first to improve the performance of the circuit system and power supply as much as possible, and then consider the signal processing technology.

Digital Fourier filtering (Digital Fourier filtering) preprocessing method can effectively filter out high-frequency noise and low-frequency noise caused by instrument background noise or baseline drift, etc., and increase the spectral signal-to-noise ratio. Digital Fourier filtering first performs fast Fourier transform (FFT) on the data, acts on the Gaussian window function in the frequency space, and then inverse fast Fourier transform (IFFT) to obtain spectral data after bandpass filtering. The mean and standard deviation of the Gaussian function determine the center frequency and bandwidth of the bandpass filter respectively, and the determination of the filtering parameters is usually realized by numerical optimization method to obtain the best filtering effect.

2.2 Spectral analysis module

The spectral analysis module uses three algorithms of Euclidean distance, neural network and support vector machine to analyze and process the Raman spectrum.

2.2.1 Euclidean distance

Euclidean distance, also known as Euclidean metric and Euclidean distance, is a commonly used definition of distance, which is the real distance between two points in m-dimensional space. The Euclidean distance in two-dimensional space is the straight-line distance between two points.

The n-dimensional Euclidean space is a point set, and each point X of it can be expressed as (x[1], x[2], ..., x[n]), where x[i] (i=1, 2, ..., n) is a real number called the i-th coordinate of X, two points A=(a[1],a[2],...,a[n]) and B=(b[1],b[ 2], . . . , b[n]), the distance d(A, B) between is defined as the following formula. d(A, B)=sqrt[∑((a[i]-b[i])^2)] (i=1, 2, . . . , n).

According to the Euclidean distance algorithm, the Euclidean distance algorithm is used to find the most similar set of Raman data between the Raman spectrum of the test cell and the existing Raman spectrum data in the database, so as to obtain information such as its cell type. In this way, there is a certain reference for the type of cells to be tested.

2.2.2 Neural network

Neural network is an algorithmic mathematical model that models the behavior characteristics of animal neural networks and performs distributed parallel information processing. This kind of network depends on the complexity of the system, and achieves the purpose of processing information by adjusting the interconnection relationship between a large number of internal nodes.

The phenotype data analysis and processing system has been re-developed according to the development kit provided by the neural network, and realizes the functions of training sample training, Raman data classification and display, and result preservation by calling the corresponding interface functions.

2.2.3 Support Vector Machine

The support vector machine method is based on the VC dimension theory of statistical learning theory and the principle of structural risk minimization, according to the complexity of the model (that is, the learning accuracy of specific training samples) and the learning ability (that is, error-free The ability to accurately identify any sample) seeks the best compromise in order to obtain the best generalization ability.

The phenotypic data analysis and processing system is developed secondaryly by using the support vector machine development kit, and realizes the functions of training sample training, Raman data classification and display, and result preservation by calling the corresponding interface functions.

In Figure 1, the basic configuration of the phenotype data analysis and processing method based on the single-cell phenotype database is: Windows XP operating system, pre-installed MySQL database.

In Figure 2, the basic hardware configuration for support vector machine identification is: a supercomputer with GPGPU (general purpose parallel processor) operating hardware, at least two CPU cores, a computing speed of at least 2Ghz, a memory of at least 2GB, and a hard disk of at least 50G . High-speed interconnection between CPU, GPGPU and storage.

Claims (4)

1. a phenotypic data analysis and processing method based on unicellular Phenotype data base, it is characterised in that Comprise the following steps:

The cell image analyzing and processing stage: by cell image information and the unknown in analyses and comparison phenotypic data storehouse Cell image data；Extract the phenotypic characteristic of unknown cell；

Data prediction: the phenotypic characteristic extracted is processed into applicable Euclidean distance algorithm, KNN algorithm, The data that algorithm of support vector machine processes；

Carry out classification analysis based on the characteristic in unicellular Phenotype data base, find out and most preferably mate cell.

Phenotypic data analyzing and processing side based on unicellular Phenotype data base the most according to claim 1 Method, it is characterised in that described data prediction comprises the following steps:

Image is carried out greyscale transformation；

Carry out image sharpening, make gray scale contrast strengthen, thus strengthen marginal information in image；

Image is carried out smothing filtering, to filter noise source；

Search the place that in image, rate of gray level is maximum, obtain the closed contour of cell image, and then extract Feature in profile.

Phenotypic data analyzing and processing side based on unicellular Phenotype data base the most according to claim 2 Method, it is characterised in that the described smothing filtering that carries out image uses Digital Fourier filter, particularly as follows:

First data are carried out fast Fourier transform, in frequency space and Gauss function effect, the most instead Fast Fourier transform, obtains the spectroscopic data after bandpass filtering.

Phenotypic data analyzing and processing side based on unicellular Phenotype data base the most according to claim 1 Method, it is characterised in that described carry out classification analysis based on the characteristic in unicellular Phenotype data base, Including training and judging two stages；

First, obtain the typical sample having recognized cell phenotype, carry out feature extraction and data prediction, obtain Obtain feature samples SVM model is trained, find the support vector in training sample, determine SVM mould Shape parameter；Then, unknown cellular informatics is input to the most parameterized SVM model it is judged.