Protein sequential, structural and functional analysis using deep learning

Abstract

This thesis introduces the application of deep learning in protein analysis. Three research questions are proposed in terms of protein sequence representation, deep learning framework design and biological background knowledge incorporation. To address these research questions, the distributed representation of protein sequences is proposed and evaluated in sequence-level, residue-level and biological interaction-level prediction tasks; a deep learning framework is proposed for any protein sequence-based residue-level prediction tasks; two multitask deep learning frameworks are applied to take advantage of the correlation between protein structural properties; and a deep transfer learning framework is proposed to incorporate the hierarchical classification systems in biology.