Prediction of protein-binding residues: dichotomy of sequence-based methods developed using structured complexes versus disordered proteins

Abstract

Motivation There are over 30 sequence-based predictors of the protein-binding residues (PBRs). They use either structure-annotated or disorder-annotated training datasets, potentially creating a dichotomy where the structure/disorder-specific models may not be able to cross-over to accurately predict the other type. Moreover, the structure-trained predictors were shown to substantially cross-predict PBRs among residues that interact with non-protein partners (nucleic acids and small ligands). We address these issues by performing first-of-its-kind comparative study of a representative collection of disorder- and structure-trained predictors using a comprehensive benchmark set with the structure- and disorder-derived annotations of PBRs (to analyze the cross-over) and the protein-, nucleic acid- and small ligand-binding proteins (to study the cross-predictions). Results Three predictors provide accurate results: SCRIBER, ANCHOR and disoRDPbind. Some of the structure-trained methods make accurate predictions on the structure-annotated proteins. Similarly, the disorder-trained predictors predict well on the disorder-annotated proteins. However, the considered predictors generally fail to cross-over, with the exception of SCRIBER. Our study also reveals that virtually all methods substantially cross-predict PBRs, except for SCRIBER for the structure-annotated proteins and disoRDPbind for the disorder-annotated proteins. We formulate a novel hybrid predictor, hybridPBRpred, that combines results produced by disoRDPbind and SCRIBER to accurately predict disorder- and structure-annotated PBRs. HybridPBRpred generates accurate results that cross-over structure- and disorder-annotated proteins and produces relatively low amount of cross-predictions, offering an accurate alternative to predict PBRs. Availability HybridPBRpred webserver, benchmark dataset and supplementary information are available at http://biomine.cs.vcu.edu/servers/hybridPBRpred/.

Comparative Assessment of Intrinsic Disorder Predictions with a Focus on Protein and Nucleic Acid-Binding Proteins

“…The result that highlights the difficulty of the current disorder predictors with the disordered protein-binding proteins motivates the development of a new generation of methods that specifically target this difficult-to-predict class of the disordered proteins. This aligns with the active research in the prediction of the protein-binding residues, which could offer useful design clues [120,121]. Furthermore, given the importance of AUC as the predictive performance metric, developers may consider explicitly optimizing their machine learning algorithms to maximize the AUC scores [122].…”

DescribePROT: database of amino acid-level protein structure and function predictions

“…The selection of the four functional predictors included in DescribePROT was informed by two observations. First, the two major classes of these predictors—ones that are trained using the intrinsically disordered AAs that bind proteins/DNA/RNA vs. ones that are trained using structured protein–protein, protein–DNA and protein–RNA complexes—were shown to provide complementary results ( 77 , 78 ). Second, multiple recent studies demonstrate that many of these methods cross-predict the three types of interacting AAs ( 11 , 30 , 31 , 79 ).…”

“…It generates three numeric propensities for protein-, DNA- and RNA-binding by disordered AAs and the corresponding three binary labels (protein/DNA/RNA binding versus non-binding) for each AAs of the input protein chain. DisoRDPbind excels through short runtime (the three types of interactions are predicted in under a second for a single protein), was ranked among the top predictors of disordered, protein-binding AAs ( 77 ), and generates low amounts of cross-predictions ( 52 , 77 ).…”

“…It outputs both the numeric propensities for protein-binding and the corresponding binary labels for each AA in the input protein sequence. Similar to DRNApred in the context of interactions with nucleic acids, this is the only method that was specifically designed to successfully minimize cross-predictions of protein binding residues ( 60 , 77 ).…”

OUP accepted manuscript

“…The LR algorithm have been successfully applied to predict DFLs in the DFLpred tool ( Meng and Kurgan, 2016 ) and was also used to predict disorder ( Fan and Kurgan, 2014 ; Peng and Kurgan, 2012 ; Peng et al , 2014 ). SVM is a popular algorithm that was extensively used to generate predictive models in closely related areas that include prediction of MoRFs ( Disfani et al , 2012 ; Fang et al , 2013 ; Malhis et al , 2016 ; Sharma et al , 2016 , 2018a,b , 2019 ; Yan et al , 2016 ) disordered protein-binding regions ( Jones and Cozzetto, 2015 ; Zhang et al , 2020 ), protein-peptide interactions ( Zhao et al , 2018 ) and IDRs ( Ishida and Kinoshita, 2007 ; Jones and Cozzetto, 2015 ; Mizianty et al , 2010 , 2013 , 2014 ; Peng et al , 2006 ; Ward et al , 2004 ). We compare these two algorithms based on the 5-fold cross validation on the training set TR166.…”