DeepKinZero: Predicting Understudied Kinases with Zero-Shot Learning

About the Talk

Presenter

Assoc. Prof. Öznur Taştan

Abstract

Protein kinases are a large family of enzymes that catalyse the phosphorylation of other proteins. Phosphorylation events regulate intracellular signalling and play central roles in cellular mechanisms.

Current computational methods for predicting phosphorylation sites rely on conventional supervised learning, which limits predictions to kinases with known sites. DeepKinZero is a zero-shot learning approach that enables predictions for understudied kinases with no prior phosphorylation data. It embeds peptide sequences and kinase representations into vector spaces and uses a compatibility function to transfer knowledge from well-characterised kinases to poorly characterised ones, potentially expanding our understanding of an important class of drug targets.

The talk was given in Turkish, with Fatma Dinçaslan.

Date: July 16th, 2019 – 18:00 (Istanbul)