Predicting Phenotype from Genomic Sequence with Deep Neural Networks

Summary

In Yu M. et al. (2016), phenotype is translated from genotype based on gene ontology, and predicted interaction scores may be influenced by errors in gene annotations or relationship between terms. As deep learning being effective in identifying complex patterns from feature-rich datasets, especially as recurrent neural networks(RNNs) such as long short term memory(LSTM) and gated recurrent unit(GRU) are capable of dealing with long-distance sequential data, predicting genetic interactions directly from DNA or amino-acid sequences using deep learning techniques would help us gain insights into underlying complex phenotypes.