ProVal: A protein‐scoring function for the selection of native and near‐native folds

Abstract

A low-resolution scoring function for the selection of native and near-native structures from a set of predicted structures for a given protein sequence has been developed. The scoring function, ProVal (Protein Validate), used several variables that describe an aspect of protein structure for which the proximity to the native structure can be assessed quantitatively. Among the parameters included are a packing estimate, surface areas, and the contact order. A partial least squares for latent variables (PLS) model was built for each candidate set of the 28 decoy sets of structures generated for 22 different proteins using the described parameters as independent variables. The C(alpha) RMS of the candidate structures versus the experimental structure was used as the dependent variable. The final generalized scoring function was an average of all models derived, ensuring that the function was not optimized for specific fold classes or method of structure generation of the candidate folds. The results show that the crystal structure was scored best in 64% of the 28 test sets and was clearly separated from the decoys in many examples. In all the other cases in which the crystal structure did not rank first, it ranked within the top 10%. Thus, although ProVal could not distinguish between predicted structures that were similar overall in fold quality due to its inherently low resolution, it can clearly be used as a primary filter to eliminate approximately 90% of fold candidates generated by current prediction methods from all-atom modeling and further evaluation. The correlation between the predicted and actual C(alpha) RMS values varies considerably between the candidate fold sets.

Hydrophobic Potential of Mean Force as a Solvation Function for Protein Structure Prediction

“…Overall, our results appear to perform better than other scoring functions published in the literature that were tested on the same or similar decoy sets. The published scoring functions can be divided into physically derived energy functions (Dominy and Brooks, 2002;Felts et al, 2002;Gatchell et al, 2000;Hsieh and Luo, 2004;Lazaridis and Karplus, 1999;Narang et al, 2006;Petrey and Honig, 2000;Zhu et al, 2003), similar in spirit to our energy function presented here, and those that are generated as a statistical scoring function based on the frequency of observations of atom or residue contacts in the PDB database, and sometimes combined with physical forces (Berglund et al, 2004;Dehouck et al, 2006;Fain et al, 2001;McConkey et al, 2003;Mukherjee et al, 2005;Shen and Sali, 2006;Wang et al, 2004). While it might seem an advantage to use knowledge-based scoring functions since they are believed to be more reliable in protein structure prediction, we found that our energy function does significantly better in both the native ranking and the native Z score, especially for the most challenging Rosetta sets, which generate $1000-2000 decoys with native-like features.…”

Introduction to Chemoinformatics in Drug Discovery – A Personal View

“…We have recently published a low-resolution scoring function ProVal [56] developed with PLS that uses a multipole representation of side chains centered on the carbon alphas and betas that can distinguish the correct structure in the midst of plausible decoy folds in a large percentage of the 28 test cases studied (Figure 1.7). For 18 of the protein sets (∼64%), the crystal structure scored best.…”

The H-factor as a novel quality metric for homology modeling

“…The quality assessment of models has been the focus of numerous studies and various algorithms exist with scoring functions based on statistical potentials [10], local side-chain and backbone interactions [11], residue environments [12], packing estimates [13], solvation energy [14], hydrogen bonding, and geometric properties [15]. In addition, the proper stereochemistry of models can be assessed by commonly used programs such as Procheck or WhatIf [16,17].…”