Deep Learning Architecture Optimization with Metaheuristic Algorithms for Predicting BRCA1/BRCA2 Pathogenicity NGS Analysis

Motivation, BRCA1 and BRCA2 are genes with tumor suppressor activity. They involved in a considerable number of biological processes. To help the biologist classification, we developed deep learning algorithm. The question when want to construct neural network is how many hidden layers neurons should use. If inputs outputs defined by problem, difficult define. Hidden that make up each layer influence performance system predictions. There different methods for finding optimal architecture. In this paper, present two packages have developed, genetic algorithm (GA) particle swarm optimization (PSO) optimize parameters predicting pathogenicity; Results, will compare results obtained algorithms. We used datasets collected from our NGS analysis train models. It represents data collection 11,875 variants. Our preliminary show PSO provided most significant architecture compared grid search GA; Conclusions, found composed 6 275 nodes an accuracy 0.98, precision 0.99, recall specificity 0.99.