Method for constructing prediction model

Abstract

The present invention provides a method of constructing a prediction model that obtains synthetic sample data by using sequencing data of a mixed positive sample and sequencing data of a negative sample. Therefore, the model construction efficiency is improved.

Description

Technical Field

[0001] The present application relates to the field of bioinformatics and to a method for constructing a prediction model. Background Art

[0002] Existing prediction models may have problems such as overfitting, poor robustness, and weak generalization ability, resulting in low sensitivity and poor accuracy of the final prediction model.

[0003] There is always a need in the art for predictive models with high sensitivity and high accuracy. Summary of the invention

[0004] To solve the above technical problems, the inventors provide a highly sensitive and accurate prediction model by using positive samples, synthetic samples and negative samples for model training. The inventors found that mixing the sequencing data of positive samples (such as late-stage / high tumor percentage positive samples) with negative samples can simulate the signals of early positive samples / low tumor percentage positive samples. At the same time, this method provides a larger number of training samples, which can ultimately achieve data enhancement. By training the machine model with enhanced data, a highly sensitive and accurate prediction model can be obtained.

[0005] On the one hand, the present invention provides a method for constructing a prediction model, which includes: obtaining sequencing data of positive samples and sequencing data of negative samples; sampling and mixing the sequencing data of the positive samples and the sequencing data of the negative samples to obtain synthetic sample data; using the positive sample and the synthetic sample data as an experimental group and using the negative sample data as a control group to train a machine learning model.

[0006] In one aspect, the present invention also provides a prediction model constructed using the method of the present invention.

[0007] On the one hand, the present invention also provides a data enhancement method for improving the performance of a prediction model, which includes: obtaining sequencing data of positive samples and sequencing data of negative samples, sampling and mixing the sequencing data of positive samples and the sequencing data of negative samples to obtain synthetic sample data, and using the synthetic sample data as training data for machine learning.

[0008] In some embodiments, the positive sample is a late positive sample.

[0009] In some embodiments, the synthetic sample data is obtained by sampling and mixing sequencing data of a positive sample with sequencing data of multiple negative samples. Preferably, the sequencing data of a positive sample is sampled and mixed with sequencing data of 2, 3, 4, 5, 6, 7, 8, 9 or 10 negative samples.

[0010] In some embodiments, a positive sample is paired with multiple negative samples.

[0011] In some embodiments, sequencing data of a positive sample is mixed with sequencing data of multiple negative samples in proportion.

[0012] In some embodiments, sequencing data of a positive sample is mixed with sequencing data of multiple negative samples according to a fixed target tumor proportion gradient.

[0013] In some embodiments, sequencing includes targeted methylation sequencing and / or whole genome sequencing (e.g., low-depth whole genome sequencing).

[0014] In some embodiments, the tumor proportion of each sample is obtained based on whole genome sequencing data analysis, and negative samples whose tumor proportion falls within the 95% quantile of negative subjects (population) are defined as negative samples, and positive samples whose tumor proportion falls outside the 95% quantile of negative subjects are defined as positive samples.

[0015] In some embodiments, the sequencing depth is 0.1X, 0.2X, 0.3X, 0.4X, 0.5X, 0.6X, 0.7X, 0.8X, 0.9X, 1.0X, 2.0X, 4.0X, 5.0X, 6.0X, 7.0X, 8.0X, 9.0X, 10.0X, 20.0X, 30.0X, 500X, 1000X, 1500X, 2000X or more depth or any depth between the foregoing values.

[0016] In some embodiments, obtaining sequencing data of a positive sample refers to obtaining sequencing data of a late positive sample.

[0017] In some embodiments, the model is a machine learning model. In some embodiments, the machine learning model is a deep learning model.

[0018] In some embodiments, the machine learning model is selected from the group consisting of: Transformer, xgboost, random forest, support vector machine, and logistic regression. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 : Model construction flowchart.

[0020] Figure 2 :ROC curves of machine learning detection models. Top: ROC curve of CRC validation set; Bottom: ROC curve of HCC validation set. DETAILED DESCRIPTION

[0021] The present invention provides a new method for constructing a prediction model. Synthetic samples are obtained by mixing sequencing data of positive samples (e.g., positive samples with a high tumor ratio in late stage) and sequencing data of negative samples to simulate positive samples with a low tumor ratio in early stage and increase the number of training samples to achieve data enhancement. By using positive samples and synthetic samples as experimental groups and negative samples as control groups for model training, this method can obtain a prediction model with high sensitivity and accuracy when the number of samples is small (especially the number of early positive samples is small).

[0022] The method of the present invention obtains synthetic sample data that can simulate early / low tumor ratio cancer by mixing sequencing data of positive samples (such as positive samples with late stage / high tumor ratio) and sequencing data of negative samples, thereby achieving data enhancement, thereby improving the prediction sensitivity and accuracy of the cancer prediction model for cancer (especially early stage cancer), and helping to improve the generalization of the model.

[0023] Throughout this specification, reference to "one embodiment," "an embodiment," "some embodiments," "an embodiment," "a specific embodiment," "a related embodiment," "an embodiment," "an additional embodiment," or "a further embodiment," or a combination thereof, means that a particular feature, structure, or characteristic described in conjunction with the embodiment is included in at least one embodiment of the present invention. Thus, the appearances of the aforementioned phrases in various places throughout this specification do not necessarily all refer to the same embodiment. Furthermore, the particular features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0024] As used herein, the terms "optional" and "optionally" include both the cases of being selected or not being selected. For example, "optionally modified" includes both the cases of being modified and not being modified.

[0025] As used herein, the terms "a," "an," and "the" are generally construed to cover both the singular and the plural.

[0026] As used herein, the term "includes" means the phrase "including (but not limited to)", and can be used interchangeably therewith. The term "comprising" used herein means the phrase "including (but not limited to)", and can be used interchangeably therewith. The technical solutions using "includes" or "comprising" in this patent can be further defined as "consisting of" or "composed of".

[0027] As used herein, the term "late positive sample" is a sample from an advanced cancer. A person skilled in the art can distinguish between early and advanced cancers according to medical knowledge. For example, in some embodiments, stage 0 and stage 1-3 cancers are considered early cancers, and stage 4 cancers are considered advanced cancers. In some embodiments, stage 0 and stage 1-2 cancers are considered early cancers, and stage 3-4 cancers are considered advanced cancers. In some embodiments, stage 0 and stage 1 are considered early cancers, and stage 2-4 are considered advanced cancers.

[0028] In some embodiments, the positive sample described herein is from a cancer sample, and the negative sample described herein is from a healthy sample.

[0029] Synthetic sample data

[0030] The synthetic sample data of the present invention is obtained by mixing the sequencing data of the positive sample with the sequencing data of the negative sample. The sequencing data of the positive sample and the sequencing data of the negative sample can be mixed according to a ratio or a fixed target tumor proportion gradient.

[0031] In some embodiments, the synthetic sample data is obtained by sampling and mixing one positive sample sequencing data with sequencing data of multiple negative samples. Preferably, in some embodiments, one positive sample sequencing data is sampled and mixed with 2, 3, 4, 5, 6, 7, 8, 9 or 10 negative samples.

[0032] In some embodiments, the one positive sample is paired with multiple negative samples. The term "paired" means that one positive sample is diluted at different gradients or ratios using the sequencing data of the same fixed negative sample.

[0033] The “mixing in proportion” mentioned herein refers to sampling and mixing the sequencing data of positive samples and the sequencing data of negative samples according to a specific preset ratio (gradient).

[0034] The "mixing according to a fixed target tumor proportion gradient" mentioned in this article refers to pre-setting the tumor proportion (target tumor proportion) gradient of the synthetic samples, and then calculating the mixing ratio of the sequencing data of a single positive sample and the sequencing data of a negative sample based on the tumor proportion of a single positive sample and the target tumor proportion.

[0035] In some embodiments, the one positive sample is mixed with a plurality of negative samples in proportion.

[0036] In some embodiments, the one positive sample is mixed with multiple negative samples according to a fixed target tumor proportion gradient.

[0037] Methods for building models

[0038] The present invention provides a method for constructing a cancer prediction model, which comprises:

[0039] Obtaining sequencing data of positive samples and sequencing data of negative samples;

[0040] The sequencing data of the positive sample and the sequencing data of the negative sample are sampled and mixed to obtain synthetic sample data;

[0041] Use positive sample and synthetic sample data as the experimental group, and negative sample data as the control group to train the machine learning model.

[0042] In some embodiments, the number of positive samples is greater than 20, greater than 30, greater than 50, greater than 200, greater than 300, greater than 400, greater than 500, greater than 600, greater than 700, greater than 800, greater than 900, greater than 1000 or more.

[0043] In some embodiments, the number of negative samples is greater than 200, greater than 300, greater than 400, greater than 500, greater than 600, greater than 700, greater than 800, greater than 900, greater than 1000, or more.

[0044] Machine Learning Models

[0045] Those skilled in the art will appreciate that any machine learning model in the art may be used in the methods herein.

[0046] In some embodiments, the machine learning model is a deep learning model.

[0047] In some embodiments, the machine learning model is a supervised learning or unsupervised learning machine learning model.

[0048] In some embodiments, the machine learning model is selected from the group consisting of: Transformer, xgboost, random forest, support vector machine, and logistic regression.

[0049] Free DNA

[0050] In some embodiments, free DNA is dispersed floating DNA obtained from an organism, generally refers to DNA present in body fluids, free outside cells or vesicles. Free DNA is well known in the art and generally refers to DNA floating freely in body fluids.

[0051] In some embodiments, the cell-free DNA is from blood or plasma. In some embodiments, the cell-free DNA is from normal cells such as peripheral blood cells.

[0052] sample

[0053] The DNA (such as free DNA) for sequencing and analysis of the present invention can be extracted from a sample. In some embodiments, the sample is a body fluid. In some embodiments, the sample is blood. In some embodiments, the sample is peripheral blood. In some embodiments, the sample is plasma, such as the plasma of peripheral blood.

[0054] In some embodiments, the volume of the sample used in the method of the present invention is 1.0, 1.5, 2, 2.5, 3, 3.5, 4, 4.5, 5, 5.5, 6, 6.5, 7, 7.5, 8, 8.5, 9, 9.5, 10, 20, 30, 50 ml or any value between the above values. In some embodiments, the volume of the sample used in the method of the present invention is at least 0.5 ml. In some embodiments, the volume of the sample used in the method of the present invention is at most 10 ml.

[0055] Samples of the present invention can all be obtained by standard techniques in the art. For example, blood samples can be obtained by standard techniques, such as using a needle and syringe. In some embodiments, the blood sample is a peripheral blood sample. Alternatively, the blood sample can be a separated portion of peripheral blood, such as a plasma sample. In some embodiments, from the sample, for example, in the nucleosomes in the sample, free DNA is extracted. In some embodiments, after obtaining the blood sample, total DNA can be extracted from the sample using standard techniques known to those skilled in the art. In some embodiments, intact cells are removed before DNA extraction, thereby only free-floating DNA is extracted. Intact cells can be removed by any method known in the art, such as as a non-limiting example, by centrifugation or by gradient separation, such as by Ficol gradient separation. A non-limiting example of DNA extraction is FlexiGene DNA test kit (QIAGEN). Standard techniques for receiving cell-free DNA extraction are known to the technician, and a non-limiting example thereof is QIAamp Circulating Nucleic Acid test kit (QIAGEN).

[0056] Library construction and sequencing

[0057] Those skilled in the art will appreciate that any suitable method in the art may be used for library construction and sequencing.

[0058] In some embodiments, the positive samples and negative samples are obtained by sequencing free DNA.

[0059] In some embodiments, the sequencing method is a whole genome sequencing method. In some embodiments, the sequencing method is a low-depth whole genome sequencing.

[0060] In some embodiments, the sequencing method is targeted methylation sequencing.

[0061] In some embodiments, the amount of free DNA used to establish the library is 0.01, 0.05, 0.1, 0.5, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 150, 200, 250, 300, 400 or 500 ng or any value therebetween. In some embodiments, the amount of free DNA used to establish the library is 10 to 100 ng.

[0062] In some embodiments, the sequencing depth is 1500X, 1000X, 500X, 300X, 200X, 150X, 100X, 50X, 40X, 30X, 20X, 10X, 9X, 8X, 7X, 6X, 5X, 4X, 3X, 2X, 1X or a depth between any of the above values. In some embodiments, the depth of sequencing is 1000X to 30X. In some embodiments, the depth of sequencing is 10X.

[0063] In some embodiments, sequencing is next generation sequencing.Next generation sequencing, also referred to as high throughput sequencing or large-scale parallel sequencing, is any sequencing method that realizes rapid high throughput sequencing of base pairs from DNA or RNA samples.In some embodiments, sequencing is high throughput sequencing.In some embodiments, sequencing is large-scale parallel sequencing.Such sequencing is well known in the art, and can include using Illumina sequencer, MGI sequencer and nanopore sequencer and ion-based semiconductor sequencer (ion torrent) as non-limiting examples.For non-limiting examples, sequencing machines such as Illumina Nextseq 500 machines can be used, and Illumina 500 / 550V2 test kits can be used for processing.In some embodiments, sequencing is whole genome sequencing.In some embodiments, only part of the genome is sequenced.

[0064] Application of the method of the present invention

[0065] In one aspect, the present invention provides uses of the methods of the present invention or models constructed therefrom in one or more of the following aspects: early diagnosis or screening of cancer, evaluation of cancer treatment regimens, detection of cancer progression, detection of cancer residues, or other applications.

[0066] The information generated by the method of the present invention is an intermediate result for diagnosis and cannot be directly used as a basis for determining whether a person has cancer.

[0067] In some embodiments, the model constructed by the method described in the present invention can be used for cancer risk assessment. In some embodiments, the method described in the present invention can be used to determine cancer progression status or cancer stage.

[0068] Below, the present invention will illustrate the beneficial effects of the present invention by examples. Those skilled in the art will recognize that these examples are illustrative and not restrictive. These examples will not limit the scope of the present invention in any way. The experimental methods described in the following examples are conventional methods unless otherwise stated; unless otherwise stated, reagents and materials are commercially available.

[0069] The experimental techniques and methods used in this example are all conventional technical methods unless otherwise specified. For example, the experimental methods without specific conditions in the following examples are usually carried out under conventional conditions or under conditions recommended by the manufacturer. The materials, reagents, etc. used in the examples can be obtained through regular commercial channels unless otherwise specified.

[0070] Example

[0071] According to the tumor proportion of free DNA, the in silico data dilution method is used to synthesize positive samples with simulated early / low tumor proportion by negative (healthy) samples for positive (cancer) samples with late / high tumor proportion, increase the data volume, perform data enhancement, create a better model training data set, and help improve model performance.

[0072] 1. First, low-pass whole genome sequencing (low pass WGS) and targeted methylation sequencing (TBS) were performed on clinical samples (colorectal cancer (CRC), hepatocellular carcinoma (HCC), and negative samples), and the tumor fraction (TF) of the samples was estimated based on the low-pass whole genome sequencing data.

[0073] 2. Due to the inaccurate estimation of the tumor proportion of low tumor proportion samples, there is also a certain level of tumor proportion in negative samples. Take the 95% quantile of the tumor proportion of negative objects as the value, and select negative samples with tumor proportion below the value and tumor samples with tumor proportion above the threshold for subsequent analysis.

[0074] 3. Sample dilution / data synthesis:

[0075] Each late positive sample was diluted using a pool of reads consisting of fixed paired n (2-10) negative samples (within the 95% tumor percentage quantile) (all TBS data)

[0076] Dilute to a fixed tumor fraction (TF) gradient:

[0077] CRC: 0.05, 0.03, 0.02, 0.01, 0.007, 0.005, 0.003.

[0078] HCC: 0.05, 0.03, 0.02, 0.01, 0.007, 0.005, 0.003.

[0079] (All samples were diluted only to a TF gradient lower than the tumor proportion of the sample itself)

[0080] Data processing:

[0081] For TBS sample data, we randomly extract sequencing data (reads) from fastq / bam files according to the calculated ratio to synthesize new sample data. We obtain sample methylation characteristics through the TBS methylation analysis process.

[0082] Use the methylation feature data of positive samples, synthetic samples, and negative samples to train the tumor diagnosis model

[0083] 4. Model Training

[0084] Table 1: CRC and HCC trained three groups of models (xgboost) for performance comparison

[0085]

[0086] Results: The performance of modeling by adding diluted samples (synthetic data) was improved compared with modeling using late samples alone, and was able to achieve similar performance to modeling using early and late samples together (Table 2, Figure 2 ).

[0087] Table 2 Performance of each model in CRC and HCC groups

[0088] Sensitivity at 95% specificity Experimental Group Control group 1 Control group 2 CRC 80.8％ 78.8％ 73.1％ HCC 82.9％ 73.1％ 85.4％

Claims

1. A method for constructing a prediction model, comprising: Determining sequencing data of positive samples and sequencing data of negative samples; The sequencing data of the positive sample and the sequencing data of the negative sample are sampled and mixed to obtain synthetic sample data; The prediction model is trained by taking the positive sample and the synthetic sample data as the experimental group and taking the negative sample data as the control group.

2. The method according to claim 1, wherein the synthetic sample data is obtained by sampling and mixing the sequencing data of a positive sample with the sequencing data of multiple negative samples.

3. The method according to claim 2, wherein the one positive sample is paired with a plurality of negative samples.

4. The method according to claim 2 or 3, wherein the sequencing data of the one positive sample is mixed with the sequencing data of multiple negative samples in proportion.

5. The method according to claim 2 or 3, wherein the sequencing data of the one positive sample is mixed with the sequencing data of multiple negative samples according to a fixed target tumor proportion gradient.

6. A method according to any of the preceding claims, wherein the sequencing comprises targeted methylation sequencing and / or whole genome sequencing (e.g., low-depth whole genome sequencing).

7. A method according to any of the preceding claims, wherein the tumor proportion of each sample is obtained based on whole genome sequencing data analysis, and negative samples whose tumor proportion falls within the 95% quantile of negative subjects are defined as negative samples, and positive samples whose tumor proportion falls outside the 95% quantile of negative subjects are defined as positive samples.

8. A method according to any of the preceding claims, wherein the sequencing depth is 0.1X, 0.2X, 0.3X, 0.4X, 0.5X, 0.6X, 0.7X, 0.8X, 0.9X, 1.0X, 2.0X, 4.0X, 5.0X, 6.0X, 7.0X, 8.0X, 9.0X, 10.0X, 20.0X, 30.0X, 500X, 1000X, 1500X or more depth or any depth between the aforementioned values.

9. The method according to any one of the preceding claims, wherein obtaining sequencing data of a positive sample refers to obtaining sequencing data of a late positive sample.

10. A prediction model constructed according to the method of any preceding claim.

Images