Predicting residue–residue contacts using random forest models

Abstract

Motivation: Protein residue-residue contact prediction can be useful in predicting protein 3D structures. Current algorithms for such a purpose leave room for improvement. Results: We develop ProC_S3, a set of Random Forest algorithmbased models, for predicting residue-residue contact maps. The models are constructed based on a collection of 1490 nonredundant, high-resolution protein structures using >1280 sequencebased features. A new amino acid residue contact propensity matrix and a new set of seven amino acid groups based on contact preference are developed and used in ProC_S3. ProC_S3 delivers a 3-fold cross-validated accuracy of 26.9% with coverage of 4.7% for top L/5 predictions (L is the number of residues in a protein) of long-range contacts (sequence separation ≥ 24). Further benchmark tests deliver an accuracy of 29.7% and coverage of 5.6% for an independent set of 329 proteins. In the recently completed Ninth Community Wide Experiment on the Critical Assessment of Techniques for Protein Structure Prediction (CASP9), ProC_S3 is ranked as No. 1, No. 3, and No. 2 accuracies in the top L/5, L/10 and best 5 predictions of long-range contacts, respectively, among 18 automatic prediction servers.

MutPred Splice: machine learning-based prediction of exonic variants that disrupt splicing

“…RF has been extensively used in bioinformatics applications, including the prediction of disease-causing mutations [8,47-49]. The popularity of RF is due in part to its simplicity with no fine-tuning of parameters required and in part to its speed of classification, which is often faster than an equivalent SVM model [50]. In this study, as we are combining multiple classification models and evaluating different training sets, this advantage of RF (limited tuning required) over SVM (tuning required) was considerable.…”

In silico identification of enhancers on the basis of a combination of transcription factor binding motif occurrences

“…Our previous work5152 has indicated that random forest and Support Vector Machine (SVM) usually demonstrate good performance with various datasets. This finding is consistent with the recently published work of Fernandez-Delgado et al 53,.…”

“…In this work, random forest was applied because it is generally more robust than SVM, which is a parameter-sensitive method and requires a long period of time to optimize parameters. The random forest package in R software was used in this study, as in our previous study52. The ntree parameter was set to 5,000, which historically has demonstrated good performance5152, and the importance was set to TRUE.…”

“…The random forest package in R software was used in this study, as in our previous study52. The ntree parameter was set to 5,000, which historically has demonstrated good performance5152, and the importance was set to TRUE. To build a robust model, the Pareto optimization rule27 was applied, which favors a good model with better performance and fewer numbers of features.…”

Evolutionary decision rules for predicting protein contact maps

“…Li et al [40] developed ProC_S3, based on a set of Random Forest algorithm based models using 1287 sequence-based features. Marks et al [43] use a global model of maximum entropy constrained by correlated mutations from multiple sequence alignments.…”