Automatic Protein Structure Determination from Sparse NMR Spectroscopy Data

Abstract

Listen

Translate article

Automated methods can accelerate the process of protein structure determination by NMR and can dramatically reduce the human expert time required. However, NMR strategies for large proteins typically require perdeuteration and site-specific labeling. The resulting datasets are sparse and ambiguous, which poses a challenge for automated structure determination. Here, we test an approach combining two computational approaches: Modeling Employing Limited Data (MELD) and the NMR Discriminative Power (DP) Score. MELD is a physics-based Bayesian approach that uses statistical inference to produce the minimum free energy ensemble of structures given a physical model and the sparse, ambiguous experimental data. This is ensemble is then filtered using the DP score, which provides a global quality metric for assessing the agreement of a structure with the experimental data. We test the combined MELD-DP approach on five proteins (from 9-23 kDa) on increasingly sparse datasets. We find that MELD-DP produces consistently good results with rich datasets with performance decreasing only slightly for the largest protein with the sparsest dataset. Preliminary results incorporating restraints using Evolutionary Couplings together with the sparse NMR data will also be discussed.