Computational methods for the prediction of protein interactions

Establishing protein interaction networks is crucial for understanding cellular operations. Detailed knowledge of the 'interactome', the full network of protein-protein interactions, in model cellular systems should provide new insights into the structure and properties of these systems. Parallel to the first massive application of experimental techniques to the determination of protein interaction networks and protein complexes, the first computational methods, based on sequence and genomic information, have emerged.     

Computational methods in protein structure prediction

This review presents the advances in protein structure prediction from the computational methods perspective. The approaches are classified into four major categories: comparative modeling, fold recognition, first principles methods that employ database information, and first principles methods without database information. Important advances along with current limitations and challenges are presented.     

Computational methods for prediction of protein-RNA interactions

Understanding the molecular mechanism of protein-RNA recognition and complex formation is a major challenge in structural biology. Unfortunately, the experimental determination of protein-RNA complexes by X-ray crystallography and nuclear magnetic resonance spectroscopy (NMR) is tedious and difficult. Alternatively, protein-RNA interactions can be predicted by computational methods. Although less accurate than experimental observations, computational predictions can be sufficiently accurate to prompt functional hypotheses and guide experiments, e.g. to identify individual amino acid or nucleotide residues. In this article we review 10 methods for predicting protein-RNA interactions, seven of which predict RNA-binding sites from protein sequences and three from structures. We also developed a meta-predictor that uses the output of top three sequence-based primary predictors to calculate a consensus prediction, which outperforms all the primary predictors. In order to fully cover the software for predicting protein-RNA interactions, we also describe five methods for protein-RNA docking. The article highlights the strengths and shortcomings of existing methods for the prediction of protein-RNA interactions and provides suggestions for their further development.     

A structural alphabet for local protein structures: improved prediction methods

Three-dimensional protein structures can be described with a library of 3D fragments that define a structural alphabet. We have previously proposed such an alphabet, composed of 16 patterns of five consecutive amino acids, called Protein Blocks (PBs). These PBs have been used to describe protein backbones and to predict local structures from protein sequences. The Q16 prediction rate reaches 40.7% with an optimization procedure. This article examines two aspects of PBs. First, we determine the effect of the enlargement of databanks on their definition. The results show that the geometrical features of the different PBs are preserved (local RMSD value equal to 0.41 A on average) and sequence-structure specificities reinforced when databanks are enlarged. Second, we improve the methods for optimizing PB predictions from sequences, revisiting the optimization procedure and exploring different local prediction strategies. Use of a statistical optimization procedure for the sequence-local structure relation improves prediction accuracy by 8% (Q16 = 48.7%). Better recognition of repetitive structures occurs without losing the prediction efficiency of the other local folds. Adding secondary structure prediction improved the accuracy of Q16 by only 1%. An entropy index (Neq), strongly related to the RMSD value of the difference between predicted PBs and true local structures, is proposed to estimate prediction quality. The Neq is linearly correlated with the Q16 prediction rate distributions, computed for a large set of proteins. An "expected" prediction rate QE16 is deduced with a mean error of 5%.     

Computational methods for protein function analysis

Two recent advances have had the greatest impact on protein function analysis so far: the complete sequences of genomes and mRNA expression level profiles. The former has spurred the development of novel techniques to study protein function: phylogenetic profiles and gene clusters. The latter has introduced a method, not based on sequence homology, that enables one to group together functionally related genes.     

Crystallographic structure refinement of Chromatium high potential iron protein at two Angstroms resolution.

The structure of Chromatium high potential iron protein (HiPIP) has been refined by semiautomatic Fo-Fc (observed minus calculated structure amplitude Fourier methods to a convential R index, R=sum of the absolute value of Fo-Fc divided by the sum of Fo, of 24.7% for a model in which bond distances and angles are constrained to standard values. Bond length and angle constraints were applied only intermittenly during the computations. At a late stage of the refinement, atomic parameters for only the Fe4S4 cluster plus the 4 associated cystein S-gamma atoms were adjusted by least squares methods and kept fixed during the rest of the refinement. The refined model consists of 625 of the 632 nonhydrogen atoms in the protein plus 75 water molecules. Seven side chain atoms could not be located in the final electron density map. A computer program rather than visual inspection was used wherever possible in the refinement: for locating water molecules, for removing water molecules that too closely approach other atoms, for deleting atoms that lay in regions of low electron density, and for evaluating the progress of refinement. Fo-Fc Fourier refinement is sufficiently economical to be applied routinely in protein crystal structure determinations. The complete HiPIP refinement required approximately 12 hours of CDC 3600 computer time and cost less than $3000 starting from a "trial structure," based upon multipe isomorphoous replacement phases, which gave an R of 43%...     

Benchmarking of structure refinement methods for protein complex models

Protein structure docking is the process in which the quaternary structure of a protein complex is predicted from individual tertiary structures of the protein subunits. Protein docking is typically performed in two main steps. The subunits are first docked while keeping them rigid to form the complex, which is then followed by structure refinement. Structure refinement is crucial for a practical use of computational protein docking models, as it is aimed for correcting conformations of interacting residues and atoms at the interface. Here, we benchmarked the performance of eight existing protein structure refinement methods in refinement of protein complex models. We show that the fraction of native contacts between subunits is by far the most straightforward metric to improve. However, backbone dependent metrics, based on the Root Mean Square Deviation proved more difficult to improve via refinement.     

Highly constrained multiple-copy refinement of protein crystal structures

In the course of refining atomic protein structures, one often encounters difficulty with molecules that are unusually flexible or otherwise disordered. We approach the problem by combining two relatively recent developments: simultaneous refinement of multiple protein conformations and highly constrained refinement. A constrained Langevin dynamics refinement is tested on two proteins: neurotrophin-3 and glutamine synthetase. The method produces closer agreement between the calculated and observed scattering amplitudes than standard, single-copy, Gaussian atomic displacement parameter refinement. This is accomplished without significantly increasing the number of fitting parameters in the model. These results suggest that loop motion in proteins within a crystal lattice can be extensive and that it is poorly modeled by isotropic Gaussian distributions for each atom. Proteins 29:426–432, 1997. © 1997 Wiley-Liss, Inc.     

Knowledge-based prediction of protein structures

We propose a knowledge-based approach to the prediction of protein structures in cases where there is no sequence-homology to proteins with known spatial structure. Using methods from Artificial Intelligence we attempt to take into account long-range interactions within the prediction process. This allows not only the assignment of secondary but also of supersecondary structure elements. In particular, the patterns used as conditions of prediction rules are generated by learning methods from information contained in the Protein Data Base. Patterns on higher levels of the protein structure hierarchy are used as constraints to reduce the combinatorial search space. These patterns may also be used to search for specified structure motifs by interactive retrieval.     

Statistical potential for assessment and prediction of protein structures

Protein structures in the Protein Data Bank provide a wealth of data about the interactions that determine the native states of proteins. Using the probability theory, we derive an atomic distance-dependent statistical potential from a sample of native structures that does not depend on any adjustable parameters (Discrete Optimized Protein Energy, or DOPE). DOPE is based on an improved reference state that corresponds to noninteracting atoms in a homogeneous sphere with the radius dependent on a sample native structure; it thus accounts for the finite and spherical shape of the native structures. The DOPE potential was extracted from a nonredundant set of 1472 crystallographic structures. We tested DOPE and five other scoring functions by the detection of the native state among six multiple target decoy sets, the correlation between the score and model error, and the identification of the most accurate non-native structure in the decoy set. For all decoy sets, DOPE is the best performing function in terms of all criteria, except for a tie in one criterion for one decoy set. To facilitate its use in various applications, such as model assessment, loop modeling, and fitting into cryo-electron microscopy mass density maps combined with comparative protein structure modeling, DOPE was incorporated into the modeling package MODELLER-8.     

Ensemble refinement of protein crystal structures: validation and application

X-ray crystallography typically uses a single set of coordinates and B factors to describe macromolecular conformations. Refinement of multiple copies of the entire structure has been previously used in specific cases as an alternative means of representing structural flexibility. Here, we systematically validate this method by using simulated diffraction data, and we find that ensemble refinement produces better representations of the distributions of atomic positions in the simulated structures than single-conformer refinements. Comparison of principal components calculated from the refined ensembles and simulations shows that concerted motions are captured locally, but that correlations dissipate over long distances. Ensemble refinement is also used on 50 experimental structures of varying resolution and leads to decreases in R(free) values, implying that improvements in the representation of flexibility observed for the simulated structures may apply to real structures. These gains are essentially independent of resolution or data-to-parameter ratio, suggesting that even structures at moderate resolution can benefit from ensemble refinement.     

Analogy-based protein structure prediction: I. A new database of spatially similar and dissimilar structures of protein domains for testing and optimizing prediction methods

The creation and analysis of the 3Dfold_test database are described. This database comprises a large set of pairs of spatially similar protein domain structures and a larger control set of “decoys,” spatially dissimilar protein structures with approximately the same size and compactness as each member of each pair. The database is available at     

Membrane protein prediction methods

We survey computational approaches that tackle membrane protein structure and function prediction. While describing the main ideas that have led to the development of the most relevant and novel methods, we also discuss pitfalls, provide practical hints and highlight the challenges that remain. The methods covered include: sequence alignment, motif search, functional residue identification, transmembrane segment and protein topology predictions, homology and ab initio modeling. In general, predictions of functional and structural features of membrane proteins are improving, although progress is hampered by the limited amount of high-resolution experimental information available. While predictions of transmembrane segments and protein topology rank among the most accurate methods in computational biology, more attention and effort will be required in the future to ameliorate database search, homology and ab initio modeling.     

Computational methods to predict protein aggregation

In most cases, protein aggregation stems from the establishment of non-native intermolecular contacts. The formation of insoluble protein aggregates is associated with many human diseases and is a major bottleneck for the industrial production of protein-based therapeutics. Strikingly, fibrillar aggregates are naturally exploited for structural scaffolding or to generate molecular switches and can be artificially engineered to build up multi-functional nanomaterials. Thus, there is a high interest in rationalizing and forecasting protein aggregation. Here, we review the available computational toolbox to predict protein aggregation propensities, identify sequential or structural aggregation-prone regions, evaluate the impact of mutations on aggregation or recognize prion-like domains. We discuss the strengths and limitations of these algorithms and how they can evolve in the next future.     

Computational methods for protein identification from mass spectrometry data

Protein identification using mass spectrometry is an indispensable computational tool in the life sciences. A dramatic increase in the use of proteomic strategies to understand the biology of living systems generates an ongoing need for more effective, efficient, and accurate computational methods for protein identification. A wide range of computational methods, each with various implementations, are available to complement different proteomic approaches. A solid knowledge of the range of algorithms available and, more critically, the accuracy and effectiveness of these techniques is essential to ensure as many of the proteins as possible, within any particular experiment, are correctly identified. Here, we undertake a systematic review of the currently available methods and algorithms for interpreting, managing, and analyzing biological data associated with protein identification. We summarize the advances in computational solutions as they have responded to corresponding advances in mass spectrometry hardware. The evolution of scoring algorithms and metrics for automated protein identification are also discussed with a focus on the relative performance of different techniques. We also consider the relative advantages and limitations of different techniques in particular biological contexts. Finally, we present our perspective on future developments in the area of computational protein identification by considering the most recent literature on new and promising approaches to the problem as well as identifying areas yet to be explored and the potential application of methods from other areas of computational biology.     

High-accuracy protein structures by combining machine-learning with physics-based refinement

Protein structure prediction has long been available as an alternative to experimental structure determination, especially via homology modeling based on templates from related sequences. Recently, models based on distance restraints from coevolutionary analysis via machine learning to have significantly expanded the ability to predict structures for sequences without templates. One such method, AlphaFold, also performs well on sequences where templates are available but without using such information directly. Here we show that combining machine-learning based models from AlphaFold with state-of-the-art physics-based refinement via molecular dynamics simulations further improves predictions to outperform any other prediction method tested during the latest round of CASP. The resulting models have highly accurate global and local structures, including high accuracy at functionally important interface residues, and they are highly suitable as initial models for crystal structure determination via molecular replacement.     

Computational approaches for the prediction of protein function in the mitochondrion

Understanding a complex biological system, such as the mitochondrion, requires the identification of the complete repertoire of proteins targeted to the organelle, the characterization of these, and finally, the elucidation of the functional and physical interactions that occur within the mitochondrion. In the last decade, significant developments have contributed to increase our understanding of the mitochondrion, and among these, computational research has played a significant role. Not only general bioinformatics tools have been applied in the context of the mitochondrion, but also some computational techniques have been specifically developed to address problems that arose from within the mitochondrial research field. In this review the contribution of bioinformatics to mitochondrial biology is addressed through a survey of current computational methods that can be applied to predict which proteins will be localized to the mitochondrion and to unravel their functional interactions. genomic context; proteome