Protein-protein Interaction Datasets Research Articles

Improving neural protein-protein interaction extraction with knowledge selection

Protein-protein interaction (PPI) extraction from published scientific literature provides additional support for precision medicine efforts. Meanwhile, knowledge bases (KBs) contain huge amounts of structured information of protein entities and their relations, which can be encoded in entity and relation embeddings to help PPI extraction. However, the prior knowledge of protein-protein pairs must be selectively used so that it is suitable for different contexts. This paper proposes a Knowledge Selection Model (KSM) to fuse the selected prior knowledge and context information for PPI extraction. Firstly, two Transformers encode the context sequence of a protein pair according to each protein embedding, respectively. Then, the two outputs are fed to a mutual attention to capture the important context features towards the protein pair. Next, the context features are used to distill the relation embedding by a knowledge selector. Finally, the selected relation embedding and the context features are concatenated for PPI extraction. Experiments on the BioCreative VI PPI dataset show that KSM achieves a new state-of-the-art performance (38.08 % F1-score) by adding knowledge selection.

Deciphering Adverse Outcome Pathway Network Linked to Bisphenol F Using Text Mining and Systems Toxicology Approaches.

Bisphenol F (BPF) is one of several Bisphenol A (BPA) substituents that is increasingly used in manufacturing industry leading to detectable human exposure. Whereas a large number of studies have been devoted to decipher BPA effects, much less is known about its substituents. To support decision making on BPF’s safety, we have developed a new computational approach to rapidly explore the available data on its toxicological effects, combining text mining and integrative systems biology, and aiming at connecting BPF to adverse outcome pathways (AOPs). We first extracted from different databases BPF-protein associations that were expanded to protein complexes using protein-protein interaction datasets. Over-representation analysis of the protein complexes allowed to identify the most relevant biological pathways putatively targeted by BPF. Then, automatic screening of scientific abstracts from literature using the text mining tool, AOP-helpFinder, combined with data integration from various sources (AOP-wiki, CompTox, etc.) and manual curation allowed us to link BPF to AOP events. Finally, we combined all the information gathered through those analyses and built a comprehensive complex framework linking BPF to an AOP network including, as adverse outcomes, various types of cancers such as breast and thyroid malignancies. These results which integrate different types of data can support regulatory assessment of the BPA substituent, BPF, and trigger new epidemiological and experimental studies.

Machine-learning techniques for the prediction of protein–protein interactions

Protein-protein interactions (PPIs) are important for the study of protein functions and pathways involved in different biological processes, as well as for understanding the cause and progression of diseases. Several high-throughput experimental techniques have been employed for the identification of PPIs in a few model organisms, but still, there is a huge gap in identifying all possible binary PPIs in an organism. Therefore, PPI prediction using machine-learning algorithms has been used in conjunction with experimental methods for discovery of novel protein interactions. The two most popular supervised machine-learning techniques used in the prediction of PPIs are support vector machines and random forest classifiers. Bayesian-probabilistic inference has also been used but mainly for the scoring of high-throughput PPI dataset confidence measures. Recently, deep-learning algorithms have been used for sequence-based prediction of PPIs. Several clustering methods such as hierarchical and k-means are useful as unsupervised machine-learning algorithms for the prediction of interacting protein pairs without explicit data labelling. In summary, machine-learning techniques have been widely used for the prediction of PPIs thus allowing experimental researchers to study cellular PPI networks.

MCC: a Multiple Consensus Clustering Framework

Consensus clustering has emerged as an important extension of the classical clustering problem. Given a set of input clusterings of a given dataset, consensus clustering aims to find a single final clustering which is a better fit in some sense than the existing clusterings. There is a significant drawback in generating a single consensus clustering since different input clusterings could differ significantly. In this paper, we develop a new framework, called Multiple Consensus Clustering (MCC), to explore multiple clustering views of a given dataset from a set of input clusterings. Instead of generating a single consensus, we propose two sets of approaches to obtain multiple consensus. One employs the meta clustering method, and the other uses a hierarchical tree structure and further applies a dynamic programming algorithm to generate a flat partition from the hierarchical tree using the modularity measure. Multiple consensuses are finally obtained by applying consensus clustering algorithms to each cluster of the partition. Extensive experimental results on 11 real-world datasets and a case study on a Protein-Protein Interaction (PPI) dataset demonstrate the effectiveness of the MCC framework.

Estimating dispensable content in the human interactome

Protein-protein interaction (PPI) networks (interactome networks) have successfully advanced our knowledge of molecular function, disease and evolution. While much progress has been made in quantifying errors and biases in experimental PPI datasets, it remains unknown what fraction of the error-free PPIs in the cell are completely dispensable, i.e., effectively neutral upon disruption. Here, we estimate dispensable content in the human interactome by calculating the fractions of PPIs disrupted by neutral and non-neutral mutations. Starting with the human reference interactome determined by experiments, we construct a human structural interactome by building homology-based three-dimensional structural models for PPIs. Next, we map common mutations from healthy individuals as well as Mendelian disease-causing mutations onto the human structural interactome, and perform structure-based calculations of how these mutations perturb the interactome. Using our predicted as well as experimentally-determined interactome perturbation patterns by common and disease mutations, we estimate that <~20% of the human interactome is completely dispensable.

An iteration method for identifying yeast essential proteins from heterogeneous network

BackgroundEssential proteins are distinctly important for an organism’s survival and development and crucial to disease analysis and drug design as well. Large-scale protein-protein interaction (PPI) data sets exist in Saccharomyces cerevisiae, which provides us with a valuable opportunity to predict identify essential proteins from PPI networks. Many network topology-based computational methods have been designed to detect essential proteins. However, these methods are limited by the completeness of available PPI data. To break out of these restraints, some computational methods have been proposed by integrating PPI networks and multi-source biological data. Despite the progress in the research of multiple data fusion, it is still challenging to improve the prediction accuracy of the computational methods.ResultsIn this paper, we design a novel iterative model for essential proteins prediction, named Randomly Walking in the Heterogeneous Network (RWHN). In RWHN, a weighted protein-protein interaction network and a domain-domain association network are constructed according to the original PPI network and the known protein-domain association network, firstly. And then, we establish a new heterogeneous matrix by combining the two constructed networks with the protein-domain association network. Based on the heterogeneous matrix, a transition probability matrix is established by normalized operation. Finally, an improved PageRank algorithm is adopted on the heterogeneous network for essential proteins prediction. In order to eliminate the influence of the false negative, information on orthologous proteins and the subcellular localization information of proteins are integrated to initialize the score vector of proteins. In RWHN, the topology, conservative and functional features of essential proteins are all taken into account in the prediction process. The experimental results show that RWHN obviously exceeds in predicting essential proteins ten other competing methods.ConclusionsWe demonstrated that integrating multi-source data into a heterogeneous network can preserve the complex relationship among multiple biological data and improve the prediction accuracy of essential proteins. RWHN, our proposed method, is effective for the prediction of essential proteins.

Knowledge-aware attention network for protein-protein interaction extraction.

Protein-protein interaction (PPI) extraction from published scientific literature provides additional support for precision medicine efforts. However, many of the current PPI extraction methods need extensive feature engineering and cannot make full use of the prior knowledge in knowledge bases (KBs). KBs contain huge amounts of structured information about entities and relationships, therefore play a pivotal role in PPI extraction. This paper proposes a knowledge-aware attention network (KAN) to fuse prior knowledge about protein-protein pairs and context information for PPI extraction. The proposed model first adopts a diagonal-disabled multi-head attention mechanism to encode context sequence along with knowledge representations learned from KBs. Then a novel multi-dimensional attention mechanism is used to select the features that can best describe the encoded context. Experiment results on the BioCreative VI PPI dataset show that the proposed approach could acquire knowledge-aware dependencies between different words in a sequence and lead to a new state-of-the-art performance.

Detection of protein complexes from multiple protein interaction networks using graph embedding.

Cellular processes are typically carried out by protein complexes rather than individual proteins. Identifying protein complexes is one of the keys to understanding principles of cellular organization and function. Also, protein complexes are a group of interacting genes underlying similar diseases, which points out the therapeutic importance of protein complexes. With the development of life science and computing science, an increasing amount of protein-protein interaction (PPI) data becomes available, which makes it possible to predict protein complexes from PPI networks. However, most PPI data produced by high-throughput experiments often has many false positive interactions and false negative edge loss, which makes it difficult to predict complexes accurately. In this paper, we present a new method, named as MEMO (Multiple network Embedding for coMplex detectiOn), to detect protein complexes. MEMO integrates multiple PPI datasets from different species into a single PPI network by using functional orthology information across multiple species and then uses a graph embedding technology to embed protein nodes of the network into continuous vector spaces, so as to quantify the relationships between nodes and better guild the protein complex detection process. Finally, it utilizes a seed-and-extend strategy to identify protein complexes from multiple PPI networks based on the similarities of their corresponding protein representations. As part of our approach, we also define a new quality measure which combines the cluster cohesiveness and cluster density to measure the likelihood of a detected protein complex being a real protein complex. Extensive experimental results demonstrate the proposed method outperforms state-of-the-art complex detection techniques.

Clustering protein-protein interaction data with spectral clustering and fuzzy random walk

Spectral Clustering is a graph clustering algorithm that makes use of eigenvector obtained from a matrix describing pairwise similarity between data points. It provides a dimensionality reduction for clustering in lower dimensions. One example of spectral clustering application is the clustering of protein-protein interaction (PPI) network. PPI networks are usually represented as a graph network with proteins and interactions as vertices and edges respectively. However, this spectral clustering only produces a hard clustering of proteins, whereas there may be some relationship between each protein clusters, and possibly multiple functionality for each proteins that has not been detected before. Fuzzy Random Walk is a fuzzy clustering method based on transition probability from a random walk on a dataset. In this paper, we combine both Spectral Clustering and Fuzzy Random Walk to cluster PPI network of protein TP53, a protein thatplays an important role in managing cell cycle, especially in tumor cell suppression. Using PPI dataset of TP53 obtained from the STRING database, we found the combined algorithm is proven to produce both robust and fuzzy clusters with each cluster explains one of TP53 protein’s functionality related to the tumor cell.

Multidimensional Integration Analysis of Autophagy-related Modules in Colorectal Cancer

Colorectal cancer (CRC) is a common malignant tumor of the digestive tract occurring in the colon, which mainly divided into adenocarcinoma, mucinous adenocarcinoma, and undifferentiated carcinoma. However, autophagy is related to the occurrence and development of various kinds of human diseases such as cancer. There is little research on the relationship between CRC and autophagy. Hence, we performed multidimensional integration analysis to systematically explore potential relationship between autophagy and CRC. Based on gene expression datasets of colon adenocarcinoma (COAD) and protein-protein interactions (PPIs), we first identified 12 autophagy-related modules in COAD using WGCNA. Then, 9 module pairs which with significantly crosstalk were deciphered, a total of 6 functional modules. Autophagy-related genes in these modules were closely related with CRC, emphasizing that the important role of autophagy-related genes in CRC, including PPP2CA and EIF4E, etc. In addition to, by integrating transcription factor (TF)-target and RNA-associated interactions, a regulation network was constructed, in which 42 TFs (including SMAD3 and TP53, etc.) and 20 miRNAs (including miR-20 and miR-30a, etc.) were identified as pivot regulators. Pivot TFs were mainly involved in cell cycle, cell proliferation and pathways in cancer. And pivot miRNAs were demonstrated associated with CRC. It suggests that these pivot regulators might be have an effect on the development of CRC by regulating autophagy. In a word, our results suggested that multidimensional integration strategy provides a novel approach to discover potential relationships between autophagy and CRC, and further improves our understanding of autophagy and tumor in human.

An Efficient Feature Extraction Technique Based on Local Coding PSSM and Multifeatures Fusion for Predicting Protein-Protein Interactions.

Background:Increasing evidence has indicated that protein-protein interactions (PPIs) play important roles in various aspects of the structural and functional organization of a cell. Thus, continuing to uncover potential PPIs is an important topic in the biomedical domain. Although various feature extraction methods with machine learning approaches have enhanced the prediction of PPIs. There remains room for improvement by developing novel and effective feature extraction methods and classifier approaches to identify PPIs.Method:In this study, we proposed a sequence-based feature extraction method called LCPSSMMF, which combined local coding position-specific scoring matrix (PSSM) with multifeatures fusion. First, we used a novel local coding method based on PSSM to build a new PSSM (CPSSM); the advantage of this method is that it incorporated global and local feature extraction, which can account for the interactions between residues in both continuous and discontinuous regions of amino acid sequences. Second, we adopted 2 different feature extraction methods (Local Average Group [LAG] and Bigram Probability [BP]) to capture multiple key feature information by employing the evolutionary information embedded in the CPSSM matrix. Finally, feature vectors were acquired by using multifeatures fusion method.Result:To evaluate the performance of the proposed feature extraction approach, we employed support vector machine (SVM) as a prediction classifier and applied this method to yeast and human PPI datasets. The prediction accuracies of LCPSSMMF were 93.43% and 90.41% on the yeast and human datasets, respectively. Moreover, we also compared the proposed method with the previous sequence-based approaches on the yeast datasets by using the same SVM classifier. The experimental results indicated that the performance of LCPSSMMF significantly exceeded that of several other state-of-the-art methods. It is proven that the LCPSSMMF approach can capture more local and global discriminatory information than almost all previous methods and can function remarkably well in identifying PPIs. To facilitate extensive research in future proteomics studies, we developed a LCPSSMMFSVM server, which is freely available for academic use at http://219.219.62.123:8888/LCPSSMMFSVM.

Feature assisted stacked attentive shortest dependency path based Bi-LSTM model for protein–protein interaction

Knowledge about protein–protein interactions is essential for understanding the biological processes such as metabolic pathways, DNA replication, and transcription etc. However, a majority of the existing Protein–Protein Interaction (PPI) systems are dependent primarily on the scientific literature, which is not yet accessible as a structured database. Thus, efficient information extraction systems are required for identifying PPI information from the large collection of biomedical texts.In this paper, we present a novel method based on attentive deep recurrent neural network, which combines multiple levels of representations exploiting word sequences and dependency path related information to identify protein–protein interaction (PPI) information from the text. We use the stacked attentive bi-directional long short term memory (Bi-LSTM) as our recurrent neural network to solve the PPI identification problem. This model leverages joint modeling of proteins and relations in a single unified framework, which is named as the ‘Attentive Shortest Dependency Path LSTM’ (Att-sdpLSTM) model. Experimentation of the proposed technique was conducted on five popular benchmark PPI datasets, namely AiMed, BioInfer, HPRD50, IEPA, and LLL. The evaluation shows the F1-score values of 93.29%, 81.68%, 78.73%, 76.25%, &83.92% on AiMed, BioInfer, HPRD50, IEPA, and LLL dataset, respectively. Comparisons with the existing systems show that our proposed approach attains state-of-the-art performance.

A protein-protein interaction map of the TNF-induced NF-\u03baB signal transduction pathway

Tumor Necrosis Factor (TNF) has a crucial role in inflammation, cell proliferation and cell death. Dysregulation of TNF receptor 1 (TNFR1)-induced Nuclear Factor-kappa B (NF-κB) signaling leads to chronic inflammation and is associated with several human inflammatory pathologies. Hence, TNF neutralization suppresses inflammation and attenuates inflammatory pathology. However, despite its beneficial effects, anti-TNF therapy suffers from efficacy issues and severe immune side effects. There is thus an urging need to identify novel targets for pharmaceutical intervention in the NF-κB signaling pathway. Here, we present a protein-protein interaction dataset of the TNFR1-induced signaling pathway. For this, we used Virotrap, a novel method for studying protein complexes without disrupting the cellular integrity, on 12 central proteins controlling NF-κB and cell death signaling, both under resting conditions as well as upon TNF stimulation. Our dataset reveals dynamic interactions in TNFR1-induced NF-κB signaling and identifies both known as well as novel interactors that may help to further unravel the molecular mechanisms steering TNF-induced inflammatory signaling and pathology.

A generalized approach to predicting protein-protein interactions between virus and host

BackgroundViral infection involves a large number of protein-protein interactions (PPIs) between virus and its host. These interactions range from the initial binding of viral coat proteins to host membrane receptor to the hijacking the host transcription machinery by viral proteins. Therefore, identifying PPIs between virus and its host helps understand the mechanism of viral infections and design antiviral drugs. Many computational methods have been developed to predict PPIs, but most of them are intended for PPIs within a species rather than PPIs across different species such as PPIs between virus and host.ResultsIn this study, we developed a prediction model of virus-host PPIs, which is applicable to new viruses and hosts. We tested the prediction model on independent datasets of virus-host PPIs, which were not used in training the model. Despite a low sequence similarity between proteins in training datasets and target proteins in test datasets, the prediction model showed a high performance comparable to the best performance of other methods for single virus-host PPIs.ConclusionsOur method will be particularly useful to find PPIs between host and new viruses for which little information is available. The program and support data are available at http://bclab.inha.ac.kr/VirusHostPPI.

Deep Neural Network Based Predictions of Protein Interactions Using Primary Sequences.

Machine learning based predictions of protein–protein interactions (PPIs) could provide valuable insights into protein functions, disease occurrence, and therapy design on a large scale. The intensive feature engineering in most of these methods makes the prediction task more tedious and trivial. The emerging deep learning technology enabling automatic feature engineering is gaining great success in various fields. However, the over-fitting and generalization of its models are not yet well investigated in most scenarios. Here, we present a deep neural network framework (DNN-PPI) for predicting PPIs using features learned automatically only from protein primary sequences. Within the framework, the sequences of two interacting proteins are sequentially fed into the encoding, embedding, convolution neural network (CNN), and long short-term memory (LSTM) neural network layers. Then, a concatenated vector of the two outputs from the previous layer is wired as the input of the fully connected neural network. Finally, the Adam optimizer is applied to learn the network weights in a back-propagation fashion. The different types of features, including semantic associations between amino acids, position-related sequence segments (motif), and their long- and short-term dependencies, are captured in the embedding, CNN and LSTM layers, respectively. When the model was trained on Pan’s human PPI dataset, it achieved a prediction accuracy of 98.78% at the Matthew’s correlation coefficient (MCC) of 97.57%. The prediction accuracies for six external datasets ranged from 92.80% to 97.89%, making them superior to those achieved with previous methods. When performed on Escherichia coli, Drosophila, and Caenorhabditis elegans datasets, DNN-PPI obtained prediction accuracies of 95.949%, 98.389%, and 98.669%, respectively. The performances in cross-species testing among the four species above coincided in their evolutionary distances. However, when testing Mus Musculus using the models from those species, they all obtained prediction accuracies of over 92.43%, which is difficult to achieve and worthy of note for further study. These results suggest that DNN-PPI has remarkable generalization and is a promising tool for identifying protein interactions.

Knowledge mining from large scale protein-protein interaction datasets at the era of big data

Knowledge mining from large scale protein-protein interaction datasets at the era of big data

Abstract 392: Reverse functional genomics: Identifying differentially regulated functional networks as predictive biomarkers of pivotal vulnerabilities in pancreatic cancer

Abstract Pancreatic Ductal Adenocarcinoma (PDAC) is an incurable disease characterized by poor survival, a dense desmoplastic stroma, and frequent activating mutations in KRAS (&amp;lt;90%) with a relatively flat mutation landscape. Despite recent advances in RNA sequencing analysis that enabled the characterization of PDAC into at least two tumor subtypes, this alone is insufficient to define more specific patterns of oncogenic dependency. As a result, there is currently no approach for stratifying these KRAS-driven tumors for targeted therapy. To inform on the more nuanced oncogenic dependencies we investigated how the deregulation of proteins localized to the extracellular face of the cell, or the “surfaceome”, plays a role in tumor maintenance and progression by utilizing an in vivo pooled screening platform in patient-derived models. A PDAC prioritized surfaceome gene set was established based on aberrant transcription in microdissected and bulk tumor datasets, functional amplification observations via the TCGA dataset, and KRAS dependency via SILAC screening in KRASG12D inducible mouse lines. Refining this initial gene set for functional relevance, through parallel loss of function shRNA screens in orthotopic and 2D conditions, we were able to successfully define an in vivo enriched PDAC functional surfaceome. To interconnect each protein of the functional surface gene set, we utilized a protein-protein interaction dataset (STRING) and built a custom CRISPR sgRNA library for the purpose of informing the downstream oncogenic PDAC dependencies. Starting from oncogenic signaling at the cell surface, we established a set of connected oncogenic modules to functionally fingerprint each PDAC model both in vivo and in vitro. Our custom algorithm leveraging essentiality to define distinct oncogenic routes has allowed insight into the shared core oncogenic dependencies within PDAC (KRAS, MYC, etc), as well as model-specific central dependencies. Notably, many of the more central model-specific nodes, like AKT, EGFR, and UBC, were druggable targets. These unique centralities have been confirmed through differential sensitivity to MK-2206, Gifitinib/Erlotinib and AUY922 respectively. Importantly, by driving the defined network from periphery to center, we associated the unique pivotal dependencies to larger functional module sets to be leveraged as potential biomarkers. The functional modules suggested by the network and the associated vulnerability match the transcriptionally defined PDAC subtypes. To validate these module-based biomarker sets, we are currently leveraging a larger PDAC xenograft cohort, as well as patient data, to quantify the predictive capacity of the differentially regulated modules in the context of variable drug dependencies. Citation Format: Johnathon L. Rose, Alessandro Carugo, Wantong Yao, Sahil Seth, Sanjana Srinivasan, Michael Peoples, Traver Hart, Timothy Heffernan, Giulio Draetta. Reverse functional genomics: Identifying differentially regulated functional networks as predictive biomarkers of pivotal vulnerabilities in pancreatic cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr 392.

Protein–protein interactions prediction based on ensemble deep neural networks

Abstract Protein–protein interactions (PPIs) are of vital importance to most biological processes. Plenty of PPIs have been identified by wet-lab experiments in the past decades, but there are still abundant uncovered PPIs. Furthermore, wet-lab experiments are expensive and limited by the adopted experimental protocols. Although various computational models have been proposed to automatically predict PPIs and provided reliable interactions for experimental verification, the problem is still far from being solved. Novel and competent models are still anticipated. In this study, a neural network based approach called EnsDNN (Ensemble Deep Neural Networks) is proposed to predict PPIs based on different representations of amino acid sequences. Particularly, EnsDNN separately uses auto covariance descriptor, local descriptor, and multi-scale continuous and discontinuous local descriptor, to represent and explore the pattern of interactions between sequentially distant and spatially close amino acid residues. It then trains deep neural networks (DNNs) with different configurations based on each descriptor. Next, EnsDNN integrates these DNNs into an ensemble predictor to leverage complimentary information of these descriptors and of DNNs, and to predict potential PPIs. EnsDNN achieves superior performance with accuracy of 95.29%, sensitivity of 95.12%, and precision of 95.45% on predicting PPIs of Saccharomyces cerevisiae. Results on other five independent PPI datasets also demonstrate that EnsDNN gets better prediction performance than other related comparing methods.

Gene expression profiles of ion channels and receptors in mouse resistance arteries: Effects of cell type, vascular bed, and age.

Receptors and ion channels of smooth muscle cells (SMCs) and endothelial cells (ECs) are integral to the regulation of vessel diameter and tissue blood flow. Physiological roles of ion channels and receptors in skeletal muscle and mesenteric arteries have been identified; however, their gene expression profiles are undefined. We tested the hypothesis that expression profiles for ion channels and receptors governing vascular reactivity vary with cell type, vascular bed, and age. Mesenteric and superior epigastric arteries were dissected from Old (24-26months) and Young (3-6months) C57BL/6J mice. ECs and SMCs were collected for analysis with custom qRT-PCR arrays to determine expression profiles of 80 ion channel and receptor genes. Bioinformatics analyses were applied to gain insight into functional interactions. We identified 68 differences in gene expression with respect to cell type, vessel type, and age. Heat maps illustrate differential expression, and distance matrices predict patterns of coexpression. Gene networks based upon protein-protein interaction datasets and KEGG pathways illustrate biological processes affected by specific differences in gene expression. Differences in gene expression profiles are most pronounced between microvascular ECs and SMCs with subtle variations between vascular beds and age groups.

An improved approach to infer protein-protein interaction based on a hierarchical vector space model

BackgroundComparing and classifying functions of gene products are important in today’s biomedical research. The semantic similarity derived from the Gene Ontology (GO) annotation has been regarded as one of the most widely used indicators for protein interaction. Among the various approaches proposed, those based on the vector space model are relatively simple, but their effectiveness is far from satisfying.ResultsWe propose a Hierarchical Vector Space Model (HVSM) for computing semantic similarity between different genes or their products, which enhances the basic vector space model by introducing the relation between GO terms. Besides the directly annotated terms, HVSM also takes their ancestors and descendants related by “is_a” and “part_of” relations into account. Moreover, HVSM introduces the concept of a Certainty Factor to calibrate the semantic similarity based on the number of terms annotated to genes. To assess the performance of our method, we applied HVSM to Homo sapiens and Saccharomyces cerevisiae protein-protein interaction datasets. Compared with TCSS, Resnik, and other classic similarity measures, HVSM achieved significant improvement for distinguishing positive from negative protein interactions. We also tested its correlation with sequence, EC, and Pfam similarity using online tool CESSM.ConclusionsHVSM showed an improvement of up to 4% compared to TCSS, 8% compared to IntelliGO, 12% compared to basic VSM, 6% compared to Resnik, 8% compared to Lin, 11% compared to Jiang, 8% compared to Schlicker, and 11% compared to SimGIC using AUC scores. CESSM test showed HVSM was comparable to SimGIC, and superior to all other similarity measures in CESSM as well as TCSS. Supplementary information and the software are available at https://github.com/kejia1215/HVSM.