Understanding the language of proteins with deep learning

By Marie Lopez, Mohamed Mounir Moussa , Ibtissem Kadri

Overview:

Level: Medium

     With the increasing availability of large-scale biological data, the applications of deep learning approaches are now expanding to several fields of healthcare and life sciences. The promise of tackling problems that could benefit the health of millions of people together with the necessity to analyse highly dimensional and highly structured data, have led to key developments in machine learning applied to diagnostic imaging, disease severity prediction and proteomics. As proteins are the machinery of life, one of the crucial challenges for computational biology is to be able to predict the function and 3D structure of proteins from their sequences of amino-acids alone. Recently, such predictions have been made possible by the use of language modeling self-supervised trainings and the use of state-of-the-art architecture like the transformer, to help closing the gap of sequence annotation. In this workshop, we will show participants how to use protein pre-trained embeddings to build a classifier of African Covid-19 strains and predict their respective 3D structures for comparison. We will provide the necessary guidance, tools and background to understand the key concepts of biology, how to use protein pre-trained embeddings and how to predict protein 3D structures from scratch using Alphafold2.

Outcome:

     At the end of the workshop, participants should have a good understanding of how to use pre-trained embeddings and de novo structure predictions to help protein characterisation as well as knowledge about real-life use-cases. On the practical side, participants: Should understand the differences between several protein embeddings and be able to compute them for any protein sequence. Are able to perform protein 3D structure predictions and compare them with the appropriate tools.

Prerequisites:

     Participants should be comfortable with programming in Python. Have a basic understanding of key concepts in machine learning and in particular, natural language processing. No prior knowledge of biology is expected or required.