Somatic Mutation Prediction in Human DNA Sequence in Absence of Matching Samples Using Artificial Neural Network

Abstract

Somatic mutation can occur at any stages of an individual which may turn towards cancer due to normal genome cell transformations. Hence, its early detection is essential for the appropriate treatment and other purposes. Generally, mutated genes are matched with the normal tissues of the donor. In addition, mutated genes are compared with the existing publicly available mutational loads. However; this computation is too costly due to appropriate matching samples and in terms of computational complexity, time consumption, memory usage etc. In these consequences, this paper proposes an efficient machine learning based approach to distinguish the somatic single nucleotide variants in absence of matching samples. Here, we have applied multilayer perceptron of artificial neural network on the standard training sets like BRCA, COAD, PAAD, KIRC, ESC, UCEC. Then the results are validated using 10-fold cross validation technique. The maximum accuracy active by the execution of the proposed scheme is 97% with f1-measure ranges from 88-97% for different cancer types which is much higher than the existing state of the art approaches.