Harnessing deep learning for SNP-based disease prediction in genomics.

Abstract

This study investigates the use of deep learning models to predict disease status from single nucleotide polymorphism (SNP) data. Eight GEO datasets were processed using a consistent pipeline involving genotype encoding, data cleaning, and multiple feature selection strategies. A variety of DL architectures-including feedforward networks, autoencoders, CNNs, and RNNs-were trained and evaluated. The novelty of this work lies in the standardized preprocessing, feature selection, and model training pipeline applied across all datasets, allowing for a direct and fair comparison of model performance. Results consistently showed that feedforward networks and autoencoders performed best across most datasets. This work offers a practical approach to applying deep learning in genomics with potential for future extensions.