Auto-Encoding Amino Acid Sequences with LSTM

Abstract

In this thesis a sequence to sequence autoencoder for amino acid sequences is constructed. The latent representation of the autoencoder is then used to classify the amino acid sequences according to their animal kingdom. The data consists of sequences from three different kingdoms, mammals, fish and birds. The thesis includes the preprocessing necessary for the data, the construction of the sequence to sequence autoencoder and the process of classification in the latent space.