Protein structure prediction remains a grand challenge, with many protein families still lacking any representative structures. MELD (Modeling Employing Limited Data) is a new method that combines physics based simulations with ambiguous, noisy and sparse information through a Bayesian inference approach. This noisy information has the potential to speed up sampling by focusing on regions that are compatible with some subsets of the data — while using all data leads to incorrect structures. Finding the lowest free energy structures in agreement with a subset of the data produces more accurate results than either simulations or the data by themselves, and is orders of magnitude faster than regular MD. The CASP (Critical Assessment of Structure Prediction) event launched 26 years ago with the aim of tracking the advances in the field. Tools like Rosetta and homology modeling have been well known in the field for their successes. Despite significant advances, there are still many protein sequences for which predictions are not possible. Blind events such as CASP (Critical Assessment of Structure Prediction) allow the chance to put new methods such as MELD to the test. CASP routinely has two main events: structure prediction from sequence and refinement of a structure. But, it also adds some other categories that are of interest. In the last CASP a big initiative was the inclusion of a data-driven structure prediction category to identify good integrative structural biology tools. During the summer of 2018 we participated in CASP13 in the categories of structure prediction, refinement and NMR-assisted predictions with some great successes that I will highlight in this talk. A big success of the MELD approach is that it still obeys detailed balance, so populations are relevant and allow us to identify native-like structures in our ensembles. A second advantage is that it is highly flexible to incorporate different sources of data. During this talk I will showcase our use of co-evolutionary data (coming from meta-genomic sequence knowledge), NMR ambiguous data and coarse physical insights (e.g. proteins make hydrophobic cores and beta strands pair up) during CASP13.

Blind protein structure prediction with physics and ambiguous data: CASP and CASP-NMR

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.