A computational model based on long short-term memory for predicting organellar genes in plastid genomes

Abstract

Gene annotation tools for the identification of gene functions are often based on similarity with reference sequences, such as those of model organisms.If sequence data for a relevant model organism are not available, it is necessary to use data for closely related organisms, but methods for identifying related organisms are computationally intensive. We propose the application of LSTM (long short-term memory) models for the automatic annotation of genes by generating a training model with sequences of same taxonomic group. The proposed method to identify unknown sequences enables annotation without reference sequences.