Protein Language Models for Mapping Structure-Function Relationships
DescriptionThe Rostlab aims to use evolutionary information (EI) combined with machine learning (ML) and artificial intelligence (AI) to predict aspects of protein function and structure from sequence. The use of EI and ML allowed a breakthrough in secondary structure prediction 30 years ago, and it has been the root for all state-of-the-art predictions of protein structure and function, including AlphaFold2. With the recent breakthrough development of protein Language Models (pLMs), it has become possible to extract information from protein sequences and transfer it to supervised learning methods for protein predictions, reaching or surpassing state-of-the-art methods without use of evolutionary information. I will present our groups developments of pLMs and new methods to predict protein structure and function, up to assessing the effects of sequence variation upon structure and towards disease. The understanding of AI and control of database bias are crucial in computational biology and can serve as a sandbox to prepare more sensitive applications of AI in society.
TimeMonday, June 2616:30 - 17:00 CEST