CN107301323B - Method for constructing classification model related to psoriasis - Google Patents

Abstract

The invention relates to the technical field of medical detection, in particular to a method for constructing a classification model related to psoriasis, which comprises the following steps: (1) selecting psoriasis susceptible sites; (2) converting the susceptible loci into input data according to different types of susceptible loci; (3) and classifying the data by using an Adaboost-SVM model. At present, relevant technologies are lacked to classify and predict psoriasis data, and only the existence of judgment sites is remained to infer the diseased situation. The invention utilizes the effective machine learning classifier SVM to classify, integrates the SVM by the adaboost frame, and improves the accuracy of the classifier. The model can integrate SNP, amino acid and type data for classification, comprehensively considers the information of each dimension, and improves the accuracy of the classification result.

Description

Method for constructing classification model related to psoriasis

Technical Field

The invention relates to the technical field of medical detection, in particular to a method for constructing a classification model related to psoriasis.

Background

Psoriasis, also known as psoriasis, is a common complex disease, and the occurrence of psoriasis has been reported to be associated with genetic factors, particularly human leukocyte antigen regions (HLA), but the sites of true association are unknown.

With the development of sequencing technology and the deepening of genome research, high-depth sequencing and accurate variation detection of MHC (major histocompatibility complex) regions of Chinese people are reported in the last year 'Nature genetics', and a plurality of psoriasis susceptibility sites are positioned in the genome association analysis. But classification and prediction models based on the susceptibility sites of HLA regions are currently lacking. Therefore, it is urgently needed to develop related classification prediction tools for performing classification prediction on data by using HLA region susceptibility sites.

Psoriasis is most significantly associated with HLA, but current technology lacks the application of HLA domain targeting. The recent accurate variation detection of the HLA region is broken through, and the susceptible sites on the HLA, which are related to psoriasis, are accurately positioned. The invention encodes the susceptible loci according to the susceptible loci and classifies the susceptible loci by using a machine learning model Adaboost, and can integrate the information of the susceptible loci found by using HLA regions. And the machine learning model is utilized to comprehensively analyze the data, so that the classification accuracy is improved, and a basis is provided for the prevention screening of psoriasis.

Disclosure of Invention

The invention aims to solve the defects of the prior art, finds biomarkers related to psoriasis based on complete coverage of MHC (major histocompatibility complex) regions, constructs a classification model of the psoriasis based on susceptibility sites independently related to HLA regions and by using SVM (support vector machine) -Adaboost, provides a method for constructing the classification model related to the psoriasis, and provides a basis for the prevention screening of the psoriasis.

The invention is realized by the following technical scheme:

1 data processing and conversion

The variation of each sample is encoded. Variation information including HLA types (C06: 02, C07: 04, DPB1 x 05:01), single nucleotide polymorphism sites (SNP sites) and amino acids (SNP31443520, B: Y33Y, B: Y91C, B: Y140S, SNP32472030) were obtained by high throughput sequencing data.

Then, for each sample, the input data required by the invention is converted according to susceptible sites. Edit distance was scored for HLA type, and SNP and amino acid were scored with 0/1. The specific method comprises the following steps: calculating the edit distance between each individual type and the susceptible type and scoring aiming at the susceptible HLA type; secondly, aiming at the SNP locus, if the mutation exists, the mutation is marked as 1, and if the mutation does not exist, the mutation is marked as 0; ③ for amino acid mutation, if the mutation exists, the mutation is marked as 1, and if the mutation does not exist, the mutation is marked as 0.

And after the scoring is finished, randomly splitting the data, splitting the data into a test set and a training set, and paying attention to the fact that the data of the test set and the data of the training set are not overlapped. When the number of samples is small, the data may be divided into 5 parts (10 parts) by the 5-fold cross method (or 10-fold cross method), and each time 1 is taken as a test set, the rest is taken as a training set.

2 classifying data by utilizing adaboost-SVM model

The invention integrates a Support Vector Machine (SVM) classifier by using an adaboost method, integrates and utilizes all susceptible site information, and improves the accuracy of data classification.

2.1 construction of the Classification model

2.1.1 sub-classification model SVM

The support vector machine model SVM is classic machine learning classification software and belongs to supervised learning. The present invention first projects data into a high dimensional space using a gaussian kernel function (equation 1).

Wherein x is any point in space, y is the selected space center, σ is the width parameter, and K (x, y) is the spatial distance from x to y.

The separation plane is then constructed in a high dimensional space using an SVM model. The partition plane construction is mainly determined by a plurality of points closest to the partition plane (as shown in fig. 1, point a is one of the closest points), and a connecting line from the closest point to the partition plane is called a support vector, and a plane when the support vector reaches the maximum is set as the partition plane, that is, the data is maximally separated by the partition plane. The method adopts an SVM model (reference website https:// www.manning.com/books/machine-learning-in-action) based on python 2.

2.1.2 Classification model integration Algorithm Adaboost

Adaboost is an integration method for improving the performance of a classifier based on errors, and an integrated result is obtained by training each sample for multiple times, repeatedly correcting the classifier through an error rate and finally integrating. The specific method comprises the following steps: samples are first given the same equal weight. The SVM is then trained on the training number set data and the error rate for the classifier is calculated (equation 2).

Error rate ═ number of correct classifications/total number of samples (equation 2)

The gaussian kernel function σ is then adjusted, followed by another SVM on the same data set. During the second training of the classifier, the weight of each sample is re-adjusted (where the weight is a multidimensional vector), wherein the next classification weight for correctly classified samples is decreased and the next weight for incorrectly classified samples is increased. That is, the final weight when the classification is correct is larger than the weight when the classification is wrong. The specific method is to calculate the weight alpha of each classifier according to the error rate.

The weights may be updated after alpha is calculated.

The classification is correct:

and (3) classification errors:

alpha is the weight of the basic classifier in the final classifier and is the error rate of the classifier; (t) represents the sequence, t represents this time, and t +1 represents the next time;D_iis the ith training sample weight.

After the weight value D is calculated, the next iteration is started. The process of training and adjusting weights is repeated until the training error rate is 0 or the number of weak classifiers reaches a specified value. The invention adopts an adaboost integrated framework based on python2 (reference website https:// www.manning.com/books/machine-learning-in-action)

3 classifying and evaluating data

After the input training set and the test set are constructed, the input training set and the test set are substituted into the constructed adaboost-SVM model for classification. The results from the classification model are compared to actual disease or lack thereof. The results were evaluated by calculating the accuracy and plotting ROC curves.

The ROC curve is a method for selecting the best signal model. The area under the ROC curve (AUC) can be calculated to determine the classification model, which is referred to table 1.

TABLE 1

The invention has the beneficial effects that:

at present, relevant technologies are lacked to classify and predict psoriasis data, and only the existence of judgment sites is remained to infer the diseased situation. The invention utilizes the effective machine learning classifier SVM to classify, integrates the SVM by the adaboost frame, and improves the accuracy of the classifier. The model can integrate SNP, amino acid and type data for classification, comprehensively considers the information of each dimension, and improves the accuracy of the classification result.

Drawings

FIG. 1 is a schematic diagram of the construction of a separation plane using an SVM model in a high dimensional space;

FIG. 2 is a ROC curve of the classification results of the training set of the present invention;

FIG. 3 is a ROC curve of the test set classification results of the present invention.

Detailed Description

For a better understanding of the present invention, the present invention will be further described with reference to the following examples and the accompanying drawings, which are illustrative of the present invention and are not to be construed as limiting thereof.

Example 1

A total of 5168 samples from psoriasis under 30 years of age were selected for the study. And classifying the constructed models of the susceptible sites by using an adaboost-SVM model based on python2 language.

1 processing and conversion of data

In this embodiment, the variation information ped and the map file of the sample are obtained through variation detection. Thereafter, HLA domain variation information was extracted from the susceptible sites (table 2). Wherein the types (1, 2, 7) are scored according to edit distance (see scoring matrix in table 3), the amino acid sites and SNP sites (3, 4, 5, 6, 8) are scored according to presence, presence is scored as 1, and absence is scored as 0.

TABLE 2 susceptible sites

TABLE 3 edit distance scoring matrix

The data list is obtained, and the data volume is 5168 cases, so 2000 cases are selected as the training set, and the rest samples are used as the test set.

Substitution model 2

The processed data are substituted into the adaboost-SVM model constructed by the method for calculation, 9 SVM classifiers are arranged in the scheme, and the value of sigma is gradually decreased from 30 to 3 from large to small.

3 obtaining the result

As shown in fig. 2 and 3, the classification error rate of the present invention is 23.9%, the training set AUC (area under ROC curve) is 0.833, and the test set AUC is 0.868, which indicates that the present invention achieves a good effect in this embodiment.

The above-mentioned embodiments are merely illustrative of the preferred embodiments of the present invention, and do not limit the scope of the present invention, and various modifications and improvements made to the technical solution of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims (3)

1. A method for constructing a classification model related to psoriasis is characterized by comprising the following steps:

(1) selecting psoriasis susceptible sites;

(2) converting the susceptible loci into input data according to different types of susceptible loci;

(3) classifying data by using an Adaboost-SVM model;

the psoriasis susceptibility site in the step (1) comprises at least one of HLA type, SNP site and amino acid;

the susceptible sites of the HLA class comprise at least one of C06: 02, C07: 04, DPB 1: 05: 01;

the SNP site and the susceptibility site of the amino acid comprise at least one of SNP31443520, B: Y33Y, B: Y91C, B: Y140S and SNP 32472030;

the transformation method in the step (2) comprises the following steps: scoring for HLA type by edit distance, scoring for SNP and amino acid by 0/1; the specific method comprises the following steps: calculating the edit distance between each individual type and the susceptible type and scoring aiming at the susceptible HLA type; secondly, aiming at the SNP locus, if the mutation exists, the mutation is marked as 1, and if the mutation does not exist, the mutation is marked as 0; ③ aiming at the amino acid mutation, if the mutation exists, the mutation is marked as 1, and if the mutation does not exist, the mutation is marked as 0;

the classification of step (3) includes the steps of:

(31) projecting data to a high-dimensional space by using a Gaussian kernel function, and constructing a separation plane by using an SVM (support vector machine) model in the high-dimensional space;

(32) the samples are endowed with the same and same weight, then the SVM is trained on the training number set data, the error rate of the classifier is calculated, weak classifiers are trained, and then the weak classifiers obtained by training are combined into a strong classifier;

(33) the data is classified and evaluated.

2. The method for constructing a classification model related to psoriasis according to claim 1, wherein the gaussian kernel function of step (31) is formulated as:

wherein x is any point in space, y is the selected space center, σ is the width parameter, and K (x, y) is the spatial distance from x to y.

3. The method of claim 1, wherein the step (33) comprises calculating an area under the ROC curve.

Images