WO2024053903A1 - Electronic nose sensor for diagnosing lung cancer, and electronic nose system for diagnosing lung cancer using same - Google Patents

Abstract

The present invention relates to an electronic nose sensor for diagnosing lung cancer, and an electronic nose system for diagnosing lung cancer using same. More specifically, by means of the technical solution, the present invention makes it possible to manufacture an electronic nose (D2 pNose) using a bacteriophage and use the electronic nose to distinguish between lung cancer patients and healthy individuals on the basis of only respiration, and identify, through a deep learning method and a neural pattern separation (NPS) analysis, a receptor contributing to pattern separation in the classification of lung cancer patients and healthy individuals.

Description

Electronic nose sensor for lung cancer diagnosis and electronic nose system for lung cancer diagnosis using the same

The present invention relates to an electronic nose sensor for lung cancer diagnosis and an electronic nose system for lung cancer diagnosis using the same. More specifically, as a means of solving the above problem, the present invention manufactures an electronic nose (D2 pNose) using bacteriophage and uses it. Therefore, it is possible to easily distinguish between lung cancer patients and healthy people just by breathing, and the receptors that contribute to pattern separation can be identified through deep learning methods and neural pattern separation (NPS) analysis for classification of lung cancer patients and healthy people.

The 1918 influenza pandemic demonstrated that respiratory disease and the resulting social chaos could spread globally. Respiratory disease caused by the H1N1 virus, which carries genes of avian origin, was transmitted by World War I veterans. Although there were many restrictions on intercontinental travel at the time, the pandemic is estimated to have infected one-third of the world's population. A characteristic of respiratory diseases is that they spread rapidly through simple contact between people. These characteristics are also revealed in the current super-spreading phenomenon of COVID-19. Current technologies are insufficient to effectively prevent the spread of respiratory diseases and the resulting social losses, and this situation is expected to continue until vaccines and treatments are developed. As the pandemic progresses, there is a growing need for rapid, accessible, and accurate diagnostic testing methods as a means of curbing the spread of respiratory diseases.

Human breath-based detection of respiratory diseases helps prevent superspreading because it allows rapid, accessible, and non-contact testing. However, classifying odors in the environment associated with various forms of life, including human breathing, is a very difficult task because it requires detecting subtle responses against a complex background. Despite these difficulties, many animals, and even humans themselves, use their sense of smell to detect and track specific objects even in complex odor environments. It may be hard to believe, but the most effective scent detection agent is not a high-tech based sensor device, but a trained K9 detection dog. K9's 220 million receptor cells and nervous system show extremely high sensitivity and selectivity for target substances in a complex environment. Inspired by K9's capabilities, artificial olfactory organs known as electronic noses (E-noses) were developed that demonstrate excellent odor-based measurement properties. However, due to engineering limitations of K9's olfactory organ, it is impossible to completely imitate its anatomical structure. Because of these limitations, there is an increasing need for new development approaches in E-noses to distinguish complex objects in environments such as human breathing.

The present invention was developed to solve the above problems. The purpose of the present invention is a sensor and detection system for lung cancer diagnosis that can easily confirm and intuitively detect lung cancer with the naked eye by immediately confirming the color change in response to the target material. is to provide.

In addition, the purpose of the present invention is to provide a sensor and detection system for lung cancer diagnosis that can be easily applied without cumbersome devices or methods of existing lung cancer detection and save time and cost by allowing the target substance to be identified by breathing the sample. .

In addition, the purpose of the present invention is to provide a sensor and detection system for lung cancer diagnosis that can be actively used for pandemic diseases such as COVID-19 by presenting a diagnostic method that is rapid, applicable in the field, and capable of portable non-face-to-face screening.

The technical problems to be solved by the invention are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art from the description below. You will be able to.

The electronic nose sensor for lung cancer diagnosis according to the present invention,

M13 bacteriophage is arranged on the base substrate,

The color changes as the M13 bacteriophage reacts with the lung cancer patient's respiratory by-products,

The M13 bacteriophage,

Glycine, Alanine, Valine, Proline, Isoleucine, Leucine, Methionine, Tryptophan, Phenylalanine, Tyrosine , Lysine, Arginine, Histidine, L-Aspartic acid, Glutamic acid, Asparagine, Glutamine, Serine, Threonine It is characterized by containing 20 amino acid sequences, including Theronine and Cysteine.

The M13 bacteriophage,

Among the 20 amino acids, it includes the amino acid sequences of Histidine, Valine, L-Aspartic acid, Glutamic acid, Serine, Glutamine, and Asparagine. It is characterized by:

The respiratory by-products of the lung cancer patient are,

4-Methyl-octane, 2-Ethyl-1-hexanol, 2-Ethyl-4-methyl-1pentanol -1pentanol), 2,3,4-trimethyl-pentane (2,3,4-Trimethyl-pentane), 2,3-Dimethyl-hexane (2,3-Dimethyl-hexane), 3-ethyl-3-methyl- 2-Ethyl-3-methyl-2-pentanone, 2-Methyl-4,6-octadiyn-3-one, 2 -Propyl-1-pentanol (2-Propyl-1-pentanol), 6,10-Dimethyl-5,9-dodecadien-2-one ) is characterized in that the color changes in response to at least one of the following.

A sensing unit provided with an electronic nose sensor in which M13 bacteriophages are arranged on a base substrate;

An optical data acquisition unit connected to the sensing unit and obtaining optical data when the electronic nose sensor reacts with an unknown substance and shows a color change; and

A determination unit connected to the optical data acquisition unit and identifying the unknown substance reacting with the electronic nose sensor as one of the target substances by comparing the optical data for the unknown substance with stored data for each target substance that has already been stored; Including,

The color of the sensing unit changes as the M13 bacteriophage reacts with the lung cancer patient's respiratory by-products,

The M13 bacteriophage,

Glycine, Alanine, Valine, Proline, Isoleucine, Leucine, Methionine, Tryptophan, Phenylalanine, Tyrosine , Lysine, Arginine, Histidine, L-Aspartic acid, Glutamic acid, Asparagine, Glutamine, Serine, Threonine It is characterized by containing 20 amino acid sequences, including Theronine and Cysteine.

The M13 bacteriophage,

Among the 20 amino acids, it includes the amino acid sequences of Histidine, Valine, L-Aspartic acid, Glutamic acid, Serine, Glutamine, and Asparagine. It is characterized by:

The respiratory by-products of the lung cancer patient are,

4-Methyl-octane, 2-Ethyl-1-hexanol, 2-Ethyl-4-methyl-1pentanol -1pentanol), 2,3,4-trimethyl-pentane (2,3,4-Trimethyl-pentane), 2,3-Dimethyl-hexane (2,3-Dimethyl-hexane), 3-ethyl-3-methyl- 2-Ethyl-3-methyl-2-pentanone, 2-Methyl-4,6-octadiyn-3-one, 2 -Propyl-1-pentanol (2-Propyl-1-pentanol), 6,10-Dimethyl-5,9-dodecadien-2-one ) is characterized in that the color changes in response to at least one of the following.

As a means of solving the above problem, the present invention relates to a sensor and detection system for lung cancer diagnosis based on M13 bacteriophage, which can immediately confirm the color change in response to the target material, making it easy to visually check and intuitively detect lung cancer. There is an effect.

In addition, the present invention can identify the target substance by breathing the sample, so it can be easily applied without the cumbersome devices or methods of existing lung cancer detection, which has the effect of saving time and cost.

In addition, the present invention provides a diagnostic method that is quick, applicable in the field, and capable of portable non-face-to-face screening, so it can be actively used for pandemic diseases such as COVID-19.

In addition, compared to the existing electronic nose, the electronic nose system of the present invention has a number of receptors of 1 million, so it is an electronic nose sensor that can be applied to in vitro diagnostic medical devices as a sensor with excellent sensitivity, and reflects an artificial intelligence model. Therefore, sensitivity and specificity can be increased due to the learning effect from various clinical trials.

Figure 1 is a schematic diagram of the M13 bacteriophage included in the electronic nose sensor for lung cancer diagnosis and the electronic nose system for lung cancer diagnosis of the present invention, which is manufactured by applying amino acids that act in mammalian olfactory receptor cells.

Figure 2 is a list of 20 amino acid sequences of M13 bacteriophage, according to an embodiment of the present invention.

Figure 3 is a schematic diagram of a phage film to which 9 amino acids optimized for lung cancer diagnosis among the 20 amino acids of M13 bacteriophage have been applied, according to an embodiment of the present invention.

Figure 4 shows the fingerprint pattern indicated by D ² pNose by classifying 61 breath samples using D ² pNose, according to an embodiment of the present invention. 31 healthy subjects (left) and D ² pNose This is a fingerprint pattern obtained by normalizing color differences when exposed to breath samples from 31 lung cancer patients (right).

Figure 5 shows the breathing of 31 healthy subjects (left) and 31 lung cancer patients (right) by classifying 61 respiratory samples using D ² pNose, according to an embodiment of the present invention. This is the average color distance.

Figure 6 is an NPS analysis for D ² pNose, according to an embodiment of the present invention, where (a) is a CSS histogram calculated for all possible combinations of 20 films, and (b) is a CSS histogram based on the CSS results. This is a histogram of sensor unit frequencies for the top 24 combinations.

Figure 7 shows the results of NPS analysis for D ² pNose according to an embodiment of the present invention. When all 20 amino acids are used in combination, the CSS is 50, and the combination of amino acids can be confirmed when the top 5 CSS results are used. . If the amino acid combination of M13 bacteriophage is Histidine, Valine, L-Aspartic acid, Glutamic acid, Serine, Glutamine, and Asparagine, the most A high CSS result value of 458 was obtained.

Figure 8 is a dendrogram based on D ² pNose composed of the best amino acid combination among the NPS analysis results for D ² pNose obtained in Figure 7, according to an embodiment of the present invention, where labels i, ii and iii are tumor These all represent patients with upper lobe and peripheral lesions (i: patient 12; ii: patient 14; iii: patient 25).

Figure 9 is a graph confirming CSS values when comparing 20 amino acids at a concentration of 10 ppm ethanol, according to an embodiment of the present invention.

Figure 10 is a color distance graph of acetone detected using D ² pNose based on an M13 bacteriophage film prepared by manipulating 20 genes, according to an embodiment of the present invention.

Figure 11 is a color distance graph of ethanol detected using D ² pNose based on an M13 bacteriophage film prepared by manipulating 20 genes, according to an embodiment of the present invention.

Figure 12 shows six VOCs (ethanol, acetone, acetaldehyde, ethyl acetate, Y-henolactone, and 2-isopropyl-4-methylthiazole), which are respiratory by-products of lung cancer patients, according to an embodiment of the present invention. ) is the Euclidean distance result of D ² pNose measured for 5 concentrations (10ppb, 50ppb, 100ppb, 500ppb, 1ppm).

Figure 13 is an electronic nose system for lung cancer diagnosis manufactured according to an embodiment of the present invention, (a) provided with a sensing unit, and (b) M13 bacteriophage color film provided with a 20 amino acid array (size: less than 1.5 and (c) a breath sampling tool.

Figure 14 shows the results of analyzing the characteristics of D ² pNose for 6 VOCs that are respiratory by-products of lung cancer patients among respiratory by-products, according to an embodiment of the present invention. (a) 6 VOCs (ethanol) using D ² pNose , acetone, acetaldehyde, ethyl acetate, Y-henolactone, and 2-isopropyl-4-methylthiazole), and (b) normalized color differences exhibited by D ² pNose in response to six VOCs. This is the fingerprint pattern obtained, and (c) is comparative data between the calculated binding energy and the experimental color distance.

Figure 15 is a dendrogram based on D ² pNose consisting of 20 amino acid combinations, according to an embodiment of the present invention, and shows measurement results for healthy subjects (blue) and lung cancer patients (red) among 62.

Figure 16 is a CT image of a lung cancer patient whose tumors were all located in the upper lobe and were peripheral lesions, according to an embodiment of the present invention.

Figure 17 is a schematic diagram (a) of the Deep Insight method for using the CNN architecture and a parallel CNN architecture (b) using the Deep Insight method according to an embodiment of the present invention.

Figure 18 is a schematic diagram showing a calculation algorithm for classification success score (CSS) according to an embodiment of the present invention.

The terms used in this specification will be briefly explained, and the present invention will be described in detail.

The terms used in the present invention are general terms that are currently widely used as much as possible while considering the function in the present invention, but this may vary depending on the intention or precedent of a person working in the art, the emergence of new technology, etc. Therefore, the terms used in the present invention should be defined based on the meaning of the term and the overall content of the present invention, rather than simply the name of the term.

When a part in the entire specification is said to “include” a certain element, this means that it does not exclude other elements but may further include other elements, unless specifically stated to the contrary.

Below, with reference to the attached drawings, embodiments of the present invention will be described in detail so that those skilled in the art can easily implement the present invention. However, the present invention may be implemented in many different forms and is not limited to the embodiments described herein.

Specific details, including the problem to be solved by the present invention, the means for solving the problem, and the effect of the invention, are included in the examples and drawings described below. The advantages and features of the present invention and methods for achieving them will become clear by referring to the embodiments described in detail below along with the accompanying drawings.

Hereinafter, the present invention will be described in more detail with reference to the attached drawings.

As shown in Figure 1, the present invention relates to a DNA-derived bacteriophage (phage) Nose (hereinafter referred to as D2 pNose) based on genetically engineered M13 phage for diagnosing respiratory diseases from human breath without pretreatment. The D2 pNose, as shown in Figure 3, is an array of phage-coated substrates with a size of 1.5 cm × 1 cm or less, and as shown in Figure 13, 20 substrates with phages of individual gene types are used. Twenty genetically engineered phages expressing the responsiveness of all mammalian olfactory receptor cells were used to produce the films.

Figure 2 shows the peptide sequence expressed in phage pVIII through genetic manipulation. The intention is to endow the device with a general and comprehensive response of the animal's olfactory system. Genetically engineered phage color films were generated using methods reported in previous work. This color film, which mimics the way skin color changes in some animal species, responds immediately (within a few milliseconds) to external substances and has been proven to be able to detect concentrations as low as 1 ppb. Breath samples for analysis can be collected following a simple procedure that does not require medical knowledge (the patient can perform it on his or her own).

In the electronic nose sensor for lung cancer diagnosis according to the present invention, M13 bacteriophages are arranged on a base substrate, and the color changes as the M13 bacteriophage reacts with respiratory by-products of lung cancer patients.

The deposition of the bacteriophage was accomplished by modifying the syringe pump. More specifically, a silicon wafer with a 100 nm gold film on a 5 nm platinum adhesive layer was selected as the substrate base for depositing M13 bacteriophages. A colorimetric sensor composed of M13 bacteriophages containing 20 types of functional peptides was fabricated using a simple pulling method.

Here, the M13 bacteriophage contains glycine, alanine, valine, proline, isoleucine, leucine, methionine, tryptophan, and phenylalanine. ), Tyrosine, Lysine, Arginine, Histidine, L-Aspartic acid, Glutamic acid, Asparagine, Glutamine, Serine It is characterized by containing 20 amino acid sequences, including serine, threonine, and cysteine.

As shown in Figure 9, the M13 bacteriophage contains histidine, valine, L-aspartic acid, glutamic acid, serine, and glutamine among the 20 amino acids. It is characterized by containing the amino acid sequences of Glutamine and Asparagine.

The respiratory by-products of lung cancer patients are 4-Methyl-octane, 2-Ethyl-1-hexanol, and 2-ethyl-4-methyl-1pentanol. (2-Ethyl-4-methyl-1pentanol), 2,3,4-trimethyl-pentane (2,3,4-Trimethyl-pentane), 2,3-dimethyl-hexane (2,3-Dimethyl-hexane), 3-Ethyl-3-methyl-2-pentanone, 2-methyl-4,6-octadiin-3-one (2-Methyl-4,6- octadiyn-3-one), 2-Propyl-1-pentanol, 6,10-dimethyl-5,9-dodecadien-2-one (6,10-Dimethyl-5 , 9-dodecadien-2-one) is characterized in that the color changes in response to at least one of the following.

The target substances that react with the electronic nose sensor for lung cancer diagnosis of the present invention are the compounds contained in the respiratory by-products of lung cancer patients described above. When normal people breathe, the above target substances are not excreted, but in lung cancer patients, the above target substances are excreted through respiration.

When the electronic nose sensor for lung cancer diagnosis of the present invention is exposed to the respiratory by-products of a patient suspected of having lung cancer, lung cancer can be diagnosed if a color change occurs in the electronic nose sensor, so it can be provided as a sensor for lung cancer diagnosis that can be intuitively judged with the naked eye. You can.

The electronic nose system for diagnosing lung cancer according to the present invention includes the electronic nose sensor for diagnosing lung cancer described above, and quantitatively analyzes the color change of the electronic nose sensor to identify which substance the unknown substance is among the target substances.

The electronic nose system for lung cancer diagnosis according to the present invention includes a sensing unit equipped with an electronic nose sensor, an optical data acquisition unit connected to the sensing unit, and a determination device connected to the optical data acquisition unit to identify the type of substance reacted in the sensing unit. Includes wealth.

The sensing unit is provided with an electronic nose sensor in which M13 bacteriophages are arranged on a base substrate. The color of the sensing unit changes as the M13 bacteriophage reacts with the lung cancer patient's respiratory by-products.

Here, the M13 bacteriophage contains glycine, alanine, valine, proline, isoleucine, leucine, methionine, tryptophan, and phenylalanine. ), Tyrosine, Lysine, Arginine, Histidine, L-Aspartic acid, Glutamic acid, Asparagine, Glutamine, Serine It is characterized by containing 20 amino acid sequences, including serine, threonine, and cysteine. As shown in Figure 9, the M13 bacteriophage contains histidine, valine, L-aspartic acid, glutamic acid, serine, and glutamine among the 20 amino acids. It is characterized by containing the amino acid sequences of Glutamine and Asparagine.

In addition, the respiratory by-products of lung cancer patients include 4-Methyl-octane, 2-Ethyl-1-hexanol, and 2-ethyl-4-methyl-1pentane. 2-Ethyl-4-methyl-1pentanol, 2,3,4-trimethyl-pentane, 2,3-dimethyl-hexane , 3-Ethyl-3-methyl-2-pentanone, 2-methyl-4,6-octadiin-3-one (2-Methyl-4,6 -octadiyn-3-one), 2-Propyl-1-pentanol, 6,10-dimethyl-5,9-dodecadien-2-one (6,10-Dimethyl- It is characterized by a color change in reaction with at least one of 5,9-dodecadien-2-one).

The optical data acquisition unit is connected to the sensing unit, and obtains optical data when the electronic nose sensor reacts with an unknown substance and shows a color change.

In one embodiment, the optical data acquisition unit is data that plots a distance vector composed of color elements (R, G, B) as a fingerprint pattern to quantify the color change generated by the electronic nose sensor reacting with an unknown substance. It could be something.

The electronic nose sensor according to the present invention changes color when the paper color film is exposed to a specific target substance, and can detect and classify based on this. Color differences are advantageous when measuring simple substances because changes can be detected with the naked eye without special equipment. However, in the case of complex substances such as human breath, it is not easy to classify existing substances based on color changes perceived with the naked eye, so the obtained color difference must be converted into numerical values and analyzed through computer-based data processing. To express this visually, a distance vector composed of color elements (R, G, B) is plotted as a fingerprint pattern, as shown in FIG. 14(b).

These color distance vectors were normalized for comparison based on the properties of the genetically engineered phages. From the fingerprint pattern, it can be seen that a unique pattern appears for each target material. Understanding how unique patterns emerge is important for designing electronic noses effectively and systematically. As shown in Figure 14(c), it can be obtained from the correlation between the reactivity of the phage color film and the phage surface chemistry.

The determination unit is connected to the optical data acquisition unit, and identifies the unknown substance reacting with the electronic nose sensor as one of the target substances by comparing optical data for the unknown substance with stored data for each target substance.

In one embodiment, the stored data for each target material may be a video image showing the changed color of the electronic nose sensor. At this time, the determination unit already has an image image of each of the target substances, and the determination unit receives the image image of the electronic nose sensor by the unknown substance, which is optical data, from the optical data acquisition unit, compares them, and determines that the unknown substance is the target substance. You can identify which of these is which one.

In one embodiment, the stored data for each target material may be a change in intensity (ΔRGB) of red, green, and blue light of the changed color of the electronic nose sensor in response to each of the target materials based on the unique color of the electronic nose sensor. At this time, the determination unit already has ΔRGB data for each of the target materials, and the determination unit converts the ΔRGB data of the unknown substance based on the optical data received from the optical data acquisition unit and compares them to determine whether the unknown substance is the target substance. You can identify which of these is which one.

In one embodiment, the stored data for each target material may be color data based on the color change of each target material. For example, color data may be data quantified in color based on ΔRGB data derived through optical data. At this time, the determination unit may further perform the step of converting the ΔRGB data of the unknown substance into color data, and compare the color data of the target substances with the color data of the unknown substance to identify the type of the unknown substance.

In contrast, the stored data for each target material may be deep learning analysis data based on the color change of each target material. Identification of the optical data in the determination unit may be based on whether the classification success score (CSS) increases compared to the stored data when assigning the measured respiratory by-products to a cluster.

For example, a deep learning approach is used to find patterns in RGB data, which can use a CNN (Convolutional Neural Network) model. Since the CNN model requires an input image, each feature vector is converted to an image using the DeepInsight method (Sharma et al., 2019), as shown in Figure 17(a), and the sample vector is converted to an image using a dimensionality reduction method that groups the sample vectors by similarity. The vector was projected onto a 2D plane. Next, the smallest rectangular boundary containing all projected sample vectors was found to map the coordinates of the projected sample vectors linearly proportionally to the image pixels. The resulting image contained 60 pixels mapped to unique features in the dataset, with the remaining pixels filled with zeros.

Afterwards, the neural pattern separation (NPS) method was applied to analyze the correlation between the selective electronic nose and human breathing characteristics using the generated CNN model (see examples below). Based on NPS, it is possible to find the surface chemistry of genetically engineered phages, which plays the most important role in distinguishing between patient's breathing and normal breathing.

Applying the NPS method to these 20 types of phage sensors is a promising way to discover receptors suitable for detecting microscopic targets in complex environments such as respiration. When performing the present invention's electronic nose system based on exposure to actual breath samples, phage amino acids were selected as the sensitive sensor that best distinguished between healthy individuals and lung cancer patients with a classification success score (CSS) calculated by NPS.

Compared to the existing electronic nose, the electronic nose system of the present invention has 1 million receptors, so it is a sensor with excellent sensitivity and can be applied to in vitro diagnostic medical devices. It reflects the artificial intelligence model and can be used in various clinical trials. Sensitivity and specificity can be increased due to learning effects.

Hereinafter, the present invention will be described in more detail through examples. The purpose, features, and advantages of the present invention will be easily understood through the following examples. The present invention is not limited to the embodiments described herein and may be embodied in other forms. The embodiments introduced here are provided to enable the idea of the present invention to be sufficiently conveyed to those skilled in the art to which the present invention pertains. Therefore, the present invention should not be limited by the following examples.

Example

1) Preparation of functional M-13 bacteriophage constructs

The major protein (pVIII) of the functional M-13 bacteriophage construct was genetically engineered to generate a DNA-grafted sensor unit in mammalian olfactory receptor cells. The primary coat protein of M13 bacteriophage was generated with 20 types of amino acids using the QuikChange II XL Site-Directed Mutagenesis Kit (Agilent Technologies) according to the manufacturer's instructions. The intended mutation was confirmed through sequencing (Bionics Co., Korea). The constructed phages were amplified using bacterial culture and purified via standard polyethylene glycol precipitation.

2) D2 pNose production

The phage deposition device was constructed by modifying a syringe pump. A silicon wafer with a 100 nm gold film on a 5 nm platinum adhesive layer was chosen as the substrate for depositing M13 bacteriophages. A colorimetric sensor composed of M13 bacteriophages containing 20 types of functional peptides was fabricated using a simple pulling method. Various phage concentrations, ion concentrations, pulling rates, and substrate surface chemistries were used for film preparation. Unless otherwise stated, gold-plated silicon substrates were used. Twenty types of colorimetric sensors were composed of red, green, and blue M13 bacteriophage films with determined pulling speeds of 20 μm/min, 30 μm/min, and 40 μm/min.

3) D2 pNose measurement analysis

To confirm the color change, pictures were taken using a handheld digital microscope (Celestron DELUXE HANDHELD DIGITAL MICROSCOPE, Torrance, CA, USA) in the gas chamber of the color sensor array set made of functional phage-based multi-array chips. After the photo was taken, it was cropped into three positions for display in the central area. The chromaticity signal of the unit cell corresponding to each sensor chip was measured as the average value of ~1000 pixel squares. The method of setting the ΔR, ΔG, and ΔB values was used as a standard for determining the amount of change in each variable. The basic characteristics of D2 pNose can be seen in FIGS. 10 to 12.

This is a color distance graph detecting acetone (FIG. 10) and ethanol (FIG. 11) using D ² pNose based on the M13 bacteriophage film produced by manipulating 20 genes.

As shown in Figure 13, D2 pNose measurement was performed at five concentrations (10ppb, 50ppb, 100ppb, 500ppb, 1ppm) of VOC (Ethanol, Acetone, Acetaldehyde, Ethyl acetate, Y-Heanolactone, and 2-Isopropyl-4-methylthiazole). was carried out in Figure 10 is the color distance of acetone detection using D2pNose, and Figure 11 is the color distance of ethanol detection using D2pNose.

Ethanol, Acetone, and Acetaldehyde were measured as molecules with relatively simple structures found in human breath, while Ethyl acetate, Y-Heanolactone, and 2-Isopropyl-4-methylthiazole were molecules with relatively complex structures found in the breath of lung cancer patients. Nine specific VOC-like molecules were found and measured. If D2 pNose has good selectivity and resolution for these VOCs, it can be assumed that it is advantageous for breath analysis for lung cancer diagnosis. Hierarchical classification analysis (HCA) was performed to discuss the selectivity of the measurement results. If the electronic nose can distinguish between VOCs despite various concentrations, it can be judged to have high selectivity. According to the analysis results, D2 pNose had the best selectivity for ethanol. In the case of other VOCs, they are classified as mixed at low concentrations, which is expected to be related to the sensitivity of D2 pNose.

4) VOC analysis of D2 pNose

Phage color films change color when exposed to specific target substances, and can be detected and classified based on this. D2 pNose consists of multiple arrays of phage color films that exhibit different reactivity toward target substances, resulting in various color changes. Previous technology reported that nine rare VOCs appeared through quantitative analysis of the breath of lung cancer patients. The unique molecular composition helps detect lung cancer through respiratory analysis.

The nine rare VOCs found only in the breath of lung cancer patients are 4-Methyl-octane, 2-Ethyl-1-hexanol, and 2-ethyl-4-methyl. -1pentanol (2-Ethyl-4-methyl-1pentanol), 2,3,4-trimethyl-pentane (2,3,4-Trimethyl-pentane), 2,3-dimethyl-hexane (2,3-Dimethyl -hexane), 3-Ethyl-3-methyl-2-pentanone, 2-methyl-4,6-octadiin-3-one (2-Methyl- 4,6-octadiyn-3-one), 2-Propyl-1-pentanol, 6,10-dimethyl-5,9-dodecadien-2-one (6,10 -Dimethyl-5,9-dodecadien-2-one). Since phage-based sensors exhibit different reactivity depending on the functional group of the molecule, the functional groups constituting the rare VOCs specified here were mainly considered. Figure 14(a) shows that various volatile organic compounds (VOC, deionized (DI) water, methanol, ethanol, propanol, diethyl ether, and hexane) at the same density (1 ppm) were selected for review purposes.

Since phage-based sensors exhibit different reactivity depending on the functional group of the molecule, the functional groups constituting the rare VOCs defined herein were mainly considered. The color difference is obtained by subtracting the initial color from the color that appears after a sufficient period of 3 minutes. It was confirmed that a unique pattern appeared for each target material.

Color differences are advantageous when measuring simple substances because changes can be detected with the naked eye without special equipment. However, in the case of complex substances such as human breath, it is difficult to classify existing substances based on color changes perceived with the naked eye. The obtained color difference must be converted into numerical values and analyzed through computer-based data processing. To express this visually, a distance vector composed of color elements (R, G, B) is plotted as a fingerprint pattern, as shown in FIG. 14(b).

These color distance vectors were normalized for comparison based on the properties of the genetically engineered phages. From the fingerprint pattern, it can be seen that a unique pattern appears for each target material. Previous research results have been confirmed that these differences in fingerprint patterns are directly related to material classification according to statistical methods. However, in the analysis of complex substances (e.g., human breath), there are limits to the application of statistical analysis due to interference between various variables.

Understanding how unique patterns emerge is important for designing electronic noses effectively and systematically. This can be ensured from the correlation between the reactivity of the phage color film and the phage surface chemistry, as shown in Figure 14(c).

If this correlation is perfect (coefficient of determination: R2 = 1), the results should line up with a trend of color distance increasing as the calculated binding energy decreases. Except for nucleic acids (R2 = 1), other VOCs are highly correlated. The R2 values are DI: 0.52, Me: 0.67, Et: 0.75, Pr: 0.72, DE: 0.59. These results demonstrated at the color film level that the chemical properties of the phage surface can be controlled by genetic manipulation. Meanwhile, there is no special relationship between the chemical properties of amino acids and their binding energies. Unlike the chemical properties of amino acids, the binding reaction with the target molecule is due to the properties of the local molecules. Even amino acids with the same chemical properties may have significantly different binding energies for specific target molecules. As a result, various reactivity can be obtained in amino acid receptors. Various reactivity plays an important role in the characteristics of electronic noses.

5) Lung cancer deep learning analysis

The presence or absence of respiratory disease can be classified by applying deep learning techniques to the pattern obtained with D2 pNose. Exhaled air from lung cancer patients and healthy subjects was collected without additional treatment. Collected breath (200 mL) was injected into a closed chamber where the D2 pNose device was located. Figures 4 and 5 show fingerprint patterns resulting from exposure of D2 pNose to human breath (left: healthy subject, right: lung cancer patient, raw data in Figure 10). The differences in breath composition between lung cancer patients and healthy subjects are subtle, and as a result, it is difficult to find obvious differences between the patterns in Figure 14(b). Therefore, a deep learning approach was used to find patterns in RGB data. Deep learning is a machine learning technology that mimics the mechanisms of the human brain through large-scale multi-layer neural networks. The large number of layers allows for a high level of abstraction, making it possible to learn representations of data collected in complex environments. In the present invention, a CNN (Convolutional Neural Network) model was used as a deep learning model. When trained according to the present invention, the CNN model achieved an accuracy of 75.15% in 5-fold cross-validation. As more data becomes available for both lung cancer patients and healthy subjects, we expect the model's accuracy to increase further.

6) Neural Pattern Separation (NPS)

Integrating D2 pNose with the NPS method is advantageous for developing receptors for detection of breathing-based respiratory diseases. Receptors that react with target substances in a vacuum (or inert matrix) exhibit a 1:1 reaction and can be easily predicted through chemical simulation. However, the composition of human breathing is very complex, and differences in this composition due to disease are minimal. Various screening effects are uninformative and cause overlapping reactions, making it difficult to find appropriate receptors. Additionally, because E-nose is composed of receptors with interactions between receptors, it is difficult to design it into an optimal configuration through simple comparison of reactivity. Therefore, applying the NPS method to these 20 types of phage sensors is a promising method to discover receptors suitable for detecting microscopic targets in complex environments such as respiration. The analysis is performed based on actual sample exposure rather than a theoretical model. The results of (hierarchical clustering analysis) for the pattern generated in D2 pNose are shown in Figure 15, and the classification success rate is 62.9%. Unlike deep learning, HCA cannot refine the diverse and complex reactivity of D2 pNose, resulting in a low classification success rate. NPS is a mechanism of the nervous system's learning and recognition process for E-nose data. Figure 9 shows the scattering of classification success scores (CSS) calculated through NPS (the calculation algorithm is in Figure 18). As shown in Figures 6 and 7, NPS identifies sensor elements that are most sensitive to differences between healthy subjects and lung cancer patients. As shown in Figure 8, the classification success rate using NPS is 87% and pattern separation is good. The selected phage genotypes (sensor elements) are histidine (His), valine (Val), aspartic acid (Asp), glutamic acid (Glu), serine (Ser), glutamine (Gln), and asparagine (Asn).

The classification results using NPS showed a good correlation with the test results of lung cancer patients. Patients 2, 4, 12, 14, 25, and 28 were misclassified as healthy subjects. Among these, patients 12, 14, and 25 had tumors located in the upper lobe and were peripheral lesions, which limited respiratory collection. These results, as shown in Figure 16, show that various characteristics are mixed even among lung cancer patients, which is an obstacle to deep learning-based classification, while only 2 healthy people (38 and 54) had lung cancer. was misclassified as a patient group. It is difficult to believe that the breathing characteristics of healthy subjects are simpler than those of lung cancer patients. Instead, the results show that selective E-nose based on NPS can clearly distinguish between the two groups through pattern separation. These results demonstrate that NPS can be successfully applied to the development of E-nose for human respiration-based measurement. Additionally, clinical outcomes for lung cancer patients show that the characteristics of collected breath samples are extremely variable. If the number of breath samples representing each feature can be sufficiently increased, learning-based verification probability can be achieved.

Experiment example

1) Experiment participants

Patients with any stage of non-small cell lung cancer (NSCLC) or small cell lung cancer (SCLC) based on the 8th edition of the American Joint Committee on Cancer Staging Manual and all enrolled patients were included in the analysis. These patients had not received cancer treatment prior to breath sampling. Healthy controls were recruited through advertisements on the Pusan National University campus and must be negative for lung cancer on chest X-ray. Information on current smoking habits and smoking history was collected and reported on a pack-year basis. Smoking cessation was defined as quitting smoking in the past 3 months. Measurements were performed in the bronchoscopy preparation room within the Department of Respiratory Medicine. Lung cancer patients (n = 31) and healthy participants (n = 31) were included in this study. The clinical characteristics of the study populations of lung cancer patients and healthy volunteers are listed in Table 1, Table 2, and Figure 11, respectively. This study was conducted under protocol no. 04-2018-035 and was performed in accordance with the Declaration of Helsinki. All patients provided informed consent before inclusion in this study. Samples analyzed in this study were prospectively collected for exhaled gas analysis.

characteristics	healthy control	lung cancer
Subject name) median age (y) Gender (male) Current smoking rate (%) Average pack (year)	31 40 13 (43.3%) 13 (43.3%) 25	31 67 21 (70%) 11 (36.6%) 42

parameter		Lung cancer (n-=31)
Tumor stage, n (%)	One 2 3 4 L.D. ED	10 (30%) 2 (6.6%) 6 (20%) 5 (16.7%) 3 (10%) 5 (16.7%)
histological subtype;	n(%) SCLC	8 (26.7%) 22 (73.3%)
Non-small cell lung cancer	COPD asthma IPF High blood pressure diabetes	5 (16.7) 2 (6.6%) 1 (3.3%) 11 (36.7%) 7 (23.3%)

2) Research design

Expiratory respiration was collected in a controlled manner from lung cancer patients and healthy subjects according to the recommendations of the European Respiratory Society technical standards (Horv'ath et al., 2017). Briefly, all participants were required to fast for at least 8 hours and rest for at least 10 minutes in a well-ventilated bronchoscopy preparation room before breath sampling. Exhaled air was collected between 9:00 and 11:00 in the morning before bronchoscopy.

Participants were instructed to breathe gently in and out through a disposable mouthpiece for 5 minutes. This mouthpiece has a particle suppression filter to prevent contamination by bacteria and viruses. Participants' lips must be covered with a mouthpiece at all times, and participants are asked to breathe only through their mouths. In this study, subjects continuously exhaled into a 200ml Tedlar sampling bag. Performing this collection does not delay routine diagnostic work. The patient did not receive individual diagnostic results from the E-nose analysis.

3) Stability and reproducibility of D2 pNose

Regarding long-term stability, the sensing part of D2 pNose is a nanofiber structure, so there is no problem with long-term physical stability. The color change of this physical structure depends only on the refractive index and the size of the nanobundles. Scratches from contact do not affect overall functionality. In terms of chemical stability, D2 pNose is not dye-based. Additionally, chemical stability to acetone was confirmed. Phage color sensors are based on the binding between target molecules and functional peptides on the phage surface. If the concentration of the target molecule is maintained in a closed space, the reaction is stable. However, because the binding energy of the molecular functional peptide is not large, the bond is easily separated when air circulates.

4) Data preparation and training of CNN using Deep Insight method

The dataset used in this work consisted of 558 samples (3 trials x 3 extraction locations x 62 subjects) with 60 unique features per sample (3 RGB values x 20 phage films). This dataset is randomly shuffled and then divided into two subsets containing 80% and 20% of all samples, respectively, for training and testing. Since the CNN model requires an image as an input, each feature vector was converted to an image using the DeepInsight method (Sharma et al., 2019), as shown in Figure 17(a). The data set was first split into 60 sample vectors, each containing the same kind of features and the same number of dimensions as the sample size (m = 558). The sample vectors were then projected onto a 2D plane using a dimensionality reduction method that groups the sample vectors by similarity. In particular, the t-SNE (t-distributed stochastic neighbor embedding) method24 was used, which determines the embedding plane by minimizing the probability of combining low-dimensional embedding and high-dimensional data. Next, the smallest rectangular boundary containing all projected sample vectors was found to map the coordinates of the projected sample vectors linearly proportionally to the image pixels. The resulting image contained 60 pixels mapped to unique features in the dataset, with the remaining pixels filled with zeros.

A simple schematic of the CNN model is shown in Figure 17(b). An input image of 20 × 20 pixels was fed into two parallel CNN structures, each with two convolutional layers and batch normalization and ReLU functions. Max pooling layer also existed after the first convolutional layer. The output of the second convolutional layer was concatenated and passed to a single fully connected layer, and the output of the fully connected layer was converted to a malignancy probability using the softmax function. To update the weights of the CNN model, we evaluated the error between the predicted and actual labels using a cross-entropy loss function with backpropagated gradients. In fact, we selected a batch of images from a randomly shuffled dataset and fed it to the CNN model at each iteration, and the Adam optimizer optimized the weights. Training and testing of the CNN model were conducted on a server equipped with a 2-core/4-thread processor (Intel Xeon, 2.5GHz), and PyTorch v1.3.1 was used to build the CNN model.

The hyperparameters of the CNN model, consisting of the number of channels and the filter size of each convolutional layer, were adjusted using Bayesian optimization techniques. Due to the small data size, the search space for the number of channels was limited to 20-60, the filter size of the first convolutional layer was limited to 1-5, and the filter size of the second convolutional layer was limited to 1. The performance of hyperparameters was evaluated using a 5-fold cross-validation method. The data set is randomly shuffled and subsequently divided into a training set and a validation set containing 4/5 and 1/5 of the total samples. The validation set was used to determine the accuracy of the trained CNN model and this procedure was repeated five times, each with a different validation set. The average accuracy of this procedure was used to find the next optimal set of hyperparameters using Bayesian optimization techniques, and the best set of hyperparameters was saved for future training. The learning rate of the Adam optimizer was also fine-tuned using the LRFinder method25.

In the training phase, we found a CNN model where the first convolutional layer had 28 channels and a filter size of 5 and 3 for the top and bottom CNN branches, respectively, and the second convolutional layer had 28 channels. The convolutional layer had 46 channels, and filter sizes of 3 and 5 for the top and bottom CNN branches, respectively, yielded the best results.

Batch sizes of 16, 32, and 64 and the number of epochs from 1 to 20 were also tested, and the model trained with a batch size of 32 images and 11 epochs performed best. The optimal learning rate of the Adam optimizer was found to be 5104. After 10 iterations of 5-fold cross-validation with this configuration, the model showed an average accuracy of 75.15 ± 1.26%.

5) NPS analysis

The signal generated from the E-nose is projected into the coding space to generate a specific pattern. The main obstacles to using E-nose as a diagnostic system based on human breathing are described below. Differences in breath composition between healthy subjects and lung cancer patients are minimal. As a result, when the electronic nose is exposed to breath samples from healthy subjects and patients, the patterns generated in the coding space overlap. The NPS method is a computational method used to transplant the pattern separation mechanism, which is the basis of the memory and learning process of the mammalian olfactory nervous system, into the E-nose. With the help of the NPS (to mimic this pattern separation), the E-nose learns to be more sensitive to subtle differences in respiratory configuration. Measurement data of D2 pNose for human breath samples were expressed as a 1 × 60 matrix (20 gene types × 3 RGB values). The 20 binary digits represent all possible configurations of the 20 film array. In each number, "1'' indicates that the sensor unit corresponding to that location is activated, and "0'' indicates that it is deactivated. When all sensor devices are activated, [1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1] is displayed. The NPS sample size for an array of n units is (2n-1). The reason for subtracting 1 is that configurations where all sensor units are disabled (all '0') are not included in the NPS analysis. HCA was performed on 1,048,575 data sets (220-1) through MATLAB code implementing the method. The den drogram in this study was obtained through HCA based on the standardized Euclidean distance and Ward's linkage method. CSS was calculated based on the cluster number output by the 'cluster' function in MATLAB.

The basic rule is that if the measured human breath sample is assigned to the correct cluster, the CSS value increases, otherwise the CSS decreases. Figure 18 shows this algorithm.

As a means of solving the above problem, the present invention relates to a sensor and detection system for lung cancer diagnosis based on M13 bacteriophage, which can immediately confirm the color change in response to the target material, making it easy to visually check and intuitively detect lung cancer. There is an effect.

In addition, the present invention can identify the target substance by breathing the sample, so it can be easily applied without the cumbersome devices or methods of existing lung cancer detection, which has the effect of saving time and cost.

In addition, the present invention provides a diagnostic method that is quick, applicable in the field, and capable of portable non-face-to-face screening, so it can be actively used for pandemic diseases such as COVID-19.

In addition, compared to the existing electronic nose, the electronic nose system of the present invention has a number of receptors of 1 million, so it is an electronic nose sensor that can be applied to in vitro diagnostic medical devices as a sensor with excellent sensitivity, and reflects an artificial intelligence model. Therefore, sensitivity and specificity can be increased due to the learning effect from various clinical trials.

As such, a person skilled in the art will understand that the technical configuration of the present invention described above can be implemented in other specific forms without changing the technical idea or essential features of the present invention.

Therefore, the embodiments described above should be understood in all respects as illustrative and not restrictive, and the scope of the present invention is indicated by the claims described later rather than the detailed description above, and the meaning and scope of the claims and their All changes or modified forms derived from the equivalent concept should be construed as falling within the scope of the present invention.

Claims (8)

1. M13 bacteriophage is arranged on the base substrate,

   The color changes as the M13 bacteriophage reacts with the lung cancer patient's respiratory by-products,

   The M13 bacteriophage,

   Glycine, Alanine, Valine, Proline, Isoleucine, Leucine, Methionine, Tryptophan, Phenylalanine, Tyrosine , Lysine, Arginine, Histidine, L-Aspartic acid, Glutamic acid, Asparagine, Glutamine, Serine, Threonine An electronic nose sensor for lung cancer diagnosis, characterized in that it contains 20 amino acid sequences of (Theronine) and Cysteine.
2. According to clause 1,

   The M13 bacteriophage,

   Among the 20 amino acids, it includes the amino acid sequences of Histidine, Valine, L-Aspartic acid, Glutamic acid, Serine, Glutamine, and Asparagine. An electronic nose sensor for lung cancer diagnosis, characterized in that:
3. According to clause 1,

   The respiratory by-products of the lung cancer patient are,

   4-Methyl-octane, 2-Ethyl-1-hexanol, 2-Ethyl-4-methyl-1pentanol -1pentanol), 2,3,4-trimethyl-pentane (2,3,4-Trimethyl-pentane), 2,3-Dimethyl-hexane (2,3-Dimethyl-hexane), 3-ethyl-3-methyl- 2-Ethyl-3-methyl-2-pentanone, 2-Methyl-4,6-octadiyn-3-one, 2 -Propyl-1-pentanol (2-Propyl-1-pentanol), 6,10-Dimethyl-5,9-dodecadien-2-one ) An electronic nose sensor for lung cancer diagnosis, characterized in that the color changes in response to at least one of the following.
4. A sensing unit provided with an electronic nose sensor in which M13 bacteriophages are arranged on a base substrate;

   An optical data acquisition unit connected to the sensing unit and obtaining optical data when the electronic nose sensor reacts with an unknown substance and shows a color change; and

   A determination unit connected to the optical data acquisition unit and identifying the unknown substance reacting with the electronic nose sensor as one of the target substances by comparing the optical data for the unknown substance with stored data for each target substance that has already been stored; Including,

   The color of the sensing unit changes as the M13 bacteriophage reacts with the unknown substance, a respiratory by-product of a lung cancer patient,

   The M13 bacteriophage,

   Glycine, Alanine, Valine, Proline, Isoleucine, Leucine, Methionine, Tryptophan, Phenylalanine, Tyrosine , Lysine, Arginine, Histidine, L-Aspartic acid, Glutamic acid, Asparagine, Glutamine, Serine, Threonine An electronic nose system for lung cancer diagnosis, characterized in that it contains 20 amino acid sequences of (Theronine) and Cysteine.
5. According to clause 4,

   The M13 bacteriophage,

   Among the 20 amino acids, it includes the amino acid sequences of Histidine, Valine, L-Aspartic acid, Glutamic acid, Serine, Glutamine, and Asparagine. An electronic nose system for lung cancer diagnosis, characterized in that:
6. According to clause 4,

   The respiratory by-products of the lung cancer patient are,

   4-Methyl-octane, 2-Ethyl-1-hexanol, 2-Ethyl-4-methyl-1pentanol -1pentanol), 2,3,4-trimethyl-pentane (2,3,4-Trimethyl-pentane), 2,3-Dimethyl-hexane (2,3-Dimethyl-hexane), 3-ethyl-3-methyl- 2-Ethyl-3-methyl-2-pentanone, 2-Methyl-4,6-octadiyn-3-one, 2 -Propyl-1-pentanol (2-Propyl-1-pentanol), 6,10-Dimethyl-5,9-dodecadien-2-one ) An electronic nose system for lung cancer diagnosis, characterized in that the color changes in response to at least one of the following.
7. According to clause 4,

   The optical data acquisition unit,

   An electronic nose system for lung cancer diagnosis, characterized in that the data is plotted as a fingerprint pattern of a distance vector composed of color elements (R, G, B) to quantify the color change generated by the electronic nose sensor reacting with an unknown substance. .
8. According to clause 4,

   Identifying the optical data in the determination unit,

   An electronic nose system for lung cancer diagnosis, characterized in that when assigning the measured respiratory by-products to a cluster, the classification success score (CSS) increases by comparing with the stored data.

Images