Genome classification with deep learning using heuristic algorithms for hyper-parameter optimization

Machine learning has been used in solving various bioinformatics tasks such as storage of biomedical data, naming conventions for new and existing organisms, gene expression, protein folding, genome classification, clustering, and sequence prediction. All these processes are needed for disease detection and study, as well as drug design and discovery. In this work, we applied deep learning models to classify genome sequences and compared their performances. We used Fully-connected neural networks, Convolutional Neural Networks, and Long-short term memory networks for 3 different classification problems. 3 preprocessing methods were tested: label encoding, one-hot encoding, and n-gram-based bag-of-words encoding. The neural network hyper-parameters are tuned with nature-inspired hierarchical algorithms called Swarm Intelligence. In the end, we reached high accuracy, increased further by the hyper-parameter optimization.