Digital Total - Computing & Data Science an der Universität Hamburg und in der Wissenschaftsmetropole Hamburg

Predicting multiple conformations of flexible proteins

85

Nicht eingeplant

20m

Von-Melle-Park 4

Poster

Beschreibung

We developed a pipeline for predicting multiple conformations of flexible proteins. A range of conformations is first generated using the deep learning model AlphaFold2 (AF2), which are then filtered using distance constraints and solvent-accessibility data from crosslinking mass spectrometry (XL-MS), using two scoring functions we developed: the crosslink and monolink probability scores (XLP, MP). The scoring functions were first benchmarked on 200 proteins (each with 300 structural decoys) using simulated XL-MS data, before being tested on an experimental test dataset. We showed that AF2 alone can only identify two out of six conformations in the test dataset, while a combination of AF2 and XLP/MP was able to identify four of six conformations, highlighting the complementarity between AF2 and XL-MS.

Keywords

protein structure prediction
computational modeling