Advances in Integrative Modeling for Proteins: Protein Loop Structure Prediction and NMR Chemical Shift Prediction
Presented By: Lichirui Zhang
Abstract:
In this talk, I will cover two projects: (1) Fast and Accurate Protein Loop Structure Prediction Using Deep Learning; and (2) Predicted and Experimental NMR Chemical Shifts at Variable Temperatures. I will first introduce a deep learning method that accurately and rapidly predicts protein loop structures. Our method, tested against the benchmark datasets CASP14 and CAMEO, demonstrates a high accuracy comparable to the state-of-the-art approach, AlphaFold2, with a substantially reduced computational cost. It also shows robust performance on challenging loop structures, even those outside of the training set. I will also describe the method's practical applications, including predicting antibody complementarity-determining regions (CDR) loop structures and refining loop structures in inexact side-chain environments. The second project, a collaborative effort, aims to elucidate the conformational origin of NMR lineshape broadening at low temperatures. Combining NMR experiments, molecular dynamics (MD) simulations, and hybrid quantum mechanics/molecular mechanics (QM/MM) calculations, we investigated the dynamics of E. coli Dihydrofolate reductase (DHFR). The chemical shifts predicted by QM/MM calculations, based on room-temperature MD simulations, show excellent agreement with the experimental low-temperature NMR spectra, offering insights into the influence of protein dynamics on NMR chemical shifts.
