Predicting protein–protein interactions based only on sequences information

Abstract

Protein-protein interactions (PPIs) are central to most biological processes. Although efforts have been devoted to the development of methodology for predicting PPIs and protein interaction networks, the application of most existing methods is limited because they need information about protein homology or the interaction marks of the protein partners. In the present work, we propose a method for PPI prediction using only the information of protein sequences. This method was developed based on a learning algorithm-support vector machine combined with a kernel function and a conjoint triad feature for describing amino acids. More than 16,000 diverse PPI pairs were used to construct the universal model. The prediction ability of our approach is better than that of other sequence-based PPI prediction methods because it is able to predict PPI networks. Different types of PPI networks have been effectively mapped with our method, suggesting that, even with only sequence information, this method could be applied to the exploration of networks for any newly discovered protein with unknown biological relativity. In addition, such supplementary experimental information can enhance the prediction ability of the method.conjoint triad ͉ support vector machine T he molecular bases of cellular operations are sustained largely by different types of interactions among proteins. Thus, a major goal of functional genomics is to determine protein interaction networks for whole organisms (1). However, only recently has it become possible to combine the traditional study of proteins as isolated entities with the analysis of large protein interaction networks by using microarray and proteomic approaches (2, 3). Such kinds of studies are significantly important because many of the functions of complex systems seem to be more closely determined by their interactions rather than by the characteristics of their individual components (4). For example, metabolic pathways, signaling cascades, and transcription control processes involve complicated interaction networks (5). Recently, interaction networks have begun to be appreciated because it is necessary to address the general principles of biological systems by means of systems biology (6). Moreover, the study of protein interaction networks has been driven by potentially practical applications in drug discovery, because it might provide great insights into mechanisms of human diseases. This study may revolutionize the pipeline of drug discovery, because drugs discovered based on the protein interaction network may specifically modulate the disease-related pathway rather than simply inhibit or activate the functions of an individual target protein (7,8). Determining accurate cellular protein interaction networks with experimental methods in combination with computational approaches therefore has become a major theme of functional genomics and proteomics efforts (9).An impressive set of experimental techniques has been developed for the systematic analysis of protein-protein interactions (...

Computational Tools for Investigating RNA-Protein Interaction Partners

“…RPISeq is a family of machine learning classifiers (RF and SVM) designed to predict the probability of interaction between a given protein and RNA. In this method, RNA sequences are encoded as normalized frequencies of RNA tetrads, and protein sequences are encoded using a conjoint triad feature (CTF) method originally proposed by Shen et al [23]. In essence, RPISeq exploits the amino acid composition of protein sequences and ribonucleotide composition of RNA sequences to predict the probability that a given pair (one protein and one RNA) will interact.…”

Domain‐mediated protein interaction prediction: From genome to network

“…Various approaches mainly differ in their encoding of sequence features and choice of learning functions. For instance, Martin et al encoded the sequence information for a protein pair by a product of signatures [80], while Shen et al proposed the use of conjoint triads, i.e., frequencies of continuous subsequences of three residues [76]. Guo et al used the auto-correlation values of seven different physicochemical scales for protein sequences as protein interaction predictors [81].…”

“…These motifs are learned from existing PPIs using only sequence data and characterize direct binding, but also could be related to protein function, which is in turn predictive of PPIs [76]. Methods based on information content analyze co-occurring subsequences of proteins with experimentally verified interactions, and use these patterns for predicting new interactions.…”

Computational Approaches to Predict Protein Interaction

“…However, considering all 20 3 amino acid triads requires an 8000-dimensional feature vector to represent a protein, which is too large for contemporary machine learning tools. Thus, the 20 amino acid types were clustered into seven groups based on their dipole strength and side chain volumes to reduce the dimensions of the feature vector (Shen, Zhang et al 2007). The frequencies of the 7 3 = 343 triads can be used to encode a protein sequence.…”

“…This led to more complicated relations than that among co-occurrence-based methods. For example, Shen et al (Shen, Zhang et al 2007) proposed to use a composition of short sequences as protein features and a following work by Chang et al (Chang, Syu et al 2010) combined these features with protein surface information. In addition to the overlap of features among different MLbased methods, they may use identical or different ML techniques.…”