2023 AIChE Annual Meeting
(27z) Assessing Prediction Fairness of AlphaFold2 in Drug Discovery
Authors
Usman Abbas - Presenter, University of Kentucky
Qing Shao, Nanjing University of Technology
Jin Chen, University of Kentucky
Xingjian Shan, University of Kentucky
AlphaFold2 is revolutionizing drug discovery by open sourcing more than 200M predicted protein structure predictions ready to use. These structure predictions enable researchers to investigate protein functions with likely conformations. Even with the claimed high accuracy, it remains unknown whether AlphaFold2 can predict the wide spectrum of protein structures equally well. In this work, we analyzed 5 million reported protein structure predictions from the AlphaFold2 database regarding model fairness. Our analysis reveals the variation of AlphaFold2âs prediction confidence with respect to residue types, secondary structures, and protein sizes. Such variation could shed light on prioritizing AlphaFold2âs predicted protein structures in drug discovery.