Identification of single nucleotide genetic polymorphism sites using machine learning methods

Abstract

The paper presents an algorithm for simulation modelling of nucleotide variations in the genomic DNA molecule. To identify single nucleotide genetic polymorphisms, it is proposed to use machine learning methods trained on simulated data. A comparative analysis of the effective classical and machine learning algorithms for identifying single nucleotide polymorphisms was performed on simulated data. The most optimal method for identifying single nucleotide genetic polymorphisms in DNA molecules at various experimental noise levels is the machine learning algorithm CART.

Identification of Single Nucleotide Genetic Polymorphism Sites Using Machine Learning Methods

“…We included in the comparative analysis two most effective existing SNP identification algorithms -the binomial distribution and entropy-based tests [2,4]. An efficient software implementation of the binomial distribution test (BDT) has been developed, a feature of which is the automation of the selection of a threshold value when identifying SNP sites.…”

“…It is proposed to use the value 10 -k as a threshold value of probabilities, where k is the average number of site coverages estimated from the simulated or experimental dataset. The published software implementation is used as an entropy-based test (EBT) [4]. Thresholds in identifying SNP sites are: the entropy E > 0,21 and the p-value < 0,5.…”

“…Complete genome sequencing or sequencing of only functionally significant regions of the human genome allows simultaneously identifying multiple sites of single nucleotide polymorphism (SNP), having diagnostic or prognostic significance for many human diseases [2,3]. Statistical methods of Fisher's exact test, binomial distribution, entropy-based tests and machine learning are used for identifying the SNPs [2,4]. These methods are quite universal and simple for program implementation, however, are computationally expensive and difficult to be effectively applied in the analysis of experimental data with a high noise level and various experimental distortions, which are sources of gaps, repetitions, and other anomalous values [1].…”