Abstract

We present a new knowledge-based Model Quality Assessment Program (MQAP) at the residue level which evaluates single protein structure models. We use a tree representation of the C α trace to train a novel Neural Network Pairwise Interaction Field (NN-PIF) to predict the global quality of a model. We also attempt to extract local quality from global quality. The model allows fast evaluation of multiple different structure models for a single sequence. In our tests on a large set of structures, our model outperforms most other methods based on different and more complex protein structure representations in both local and global quality prediction. The method is available upon request from the authors. Method-specific rankers may also built by the authors upon request.