Machine Learning Classification of Response to Internet-based Cognitive-Behavioural Therapy using Genome-Wide Association Study Data

Abstract: Genome-Wide Association Study (GWAS) data is used to predict clinical outcome of Internet-based Cognitive-Behavioural Therapy for patients suffering from depression. The original data has a very small sample size, but a huge number of features. We reduce the number of Single Nucleotide Polymorphisms (SNPs) by selecting the ones associated with unipolar depression. We define and train a Convolutional Neural Network model with the new data containing only the selected SNPs. For comparison, we also train a logistic regression model with the new data and train both models with a same size data set containing SNPs randomly chosen from the total set. The results show that the selected SNPs have stronger prediction power than the random SNPs, the trained models with the selected SNPs have better performance than a nondiscriminating classifier; however, the CNN model does not perform better than the logistic regression model. These results are discussed, with suggestions for future improvements, such as means to increase the sample size and to reduce the feature size.