Learning representations for molecular sequences

This thesis explores the utility of representation learning for bioinformatics applications. It proposes approaches for generating low dimensional embeddings for molecular sequences, which proved effective for downstream bioinformatics tasks such as protein classification and protein-protein interaction prediction. One specific theme of this thesis is to develop a scalable and computationally effective solution for large scale sequence comparisons and two successful approaches – one based on a hierarchy of models, the other on a hybrid of two methods – are presented. The representation learning approaches proposed in this thesis are generic and can be adapted for similar problems within bioinformatics and other domains