September 7, 2021 at 11:00 AM EDT - ISCBacademy COSI Series: 3D SIG by Mohammed AlQuraishi, Columbia University - Hosted by 3D SIG
AlphaFold2 burst on the life sciences stage in late 2020 with the remarkable claim that protein structure prediction has been solved. In this talk I will argue that in some fundamental sense the core scientific problem of static structure prediction is finished, but that further maturation is necessary before AlphaFold2 and similar systems can address biological questions beyond those of structure determination itself. I will outline some of these necessary developments and highlight one in particular: the prediction of structure from individual protein sequences. I will describe present challenges and opportunities, and our efforts to tackle them by combining advances in protein language modeling with end-to-end differentiable structure prediction, presenting new results on the prediction of orphan and de novo designed proteins. Time permitting, I will end by speculating on what abundant availability of structural information might mean for the future of biology.