Explaining protein-protein interactions with knowledge graph-based semantic similarity.

The application of artificial intelligence and machine learning methods for several biomedical applications, such as protein-protein interaction prediction, has gained significant traction in recent decades. However, explainability is a key aspect of using machine learning as a tool for scientific discovery. Explainable artificial intelligence approaches help clarify algorithmic mechanisms and identify potential bias in the data. Given the complexity of the biomedical domain, explanations should be grounded in domain knowledge which can be achieved by using ontologies and knowledge graphs. These knowledge graphs express knowledge about a domain by capturing different perspectives of the representation of real-world entities. However, the most popular way to explore knowledge graphs with machine learning is through using embeddings, which are not explainable. As an alternative, knowledge graph-based semantic similarity offers the advantage of being explainable. Additionally, similarity can be computed to capture different semantic aspects within the knowledge graph and increasing the explainability of predictive approaches. We propose a novel method to generate explainable vector representations, KGsim2vec, that uses aspect-oriented semantic similarity features to represent pairs of entities in a knowledge graph. Our approach employs a set of machine learning models, including decision trees, genetic programming, random forest and eXtreme gradient boosting, to predict relations between entities. The experiments reveal that considering multiple semantic aspects when representing the similarity between two entities improves explainability and predictive performance. KGsim2vec performs better than black-box methods based on knowledge graph embeddings or graph neural networks. Moreover, KGsim2vec produces global models that can capture biological phenomena and elucidate data biases.

Multi-domain knowledge graph embeddings for gene-disease association prediction.

BACKGROUND: Predicting gene-disease associations typically requires exploring diverse sources of information as well as sophisticated computational approaches. Knowledge graph embeddings can help tackle these challenges by creating representations of genes and diseases based on the scientific knowledge described in ontologies, which can then be explored by machine learning algorithms. However, state-of-the-art knowledge graph embeddings are produced over a single ontology or multiple but disconnected ones, ignoring the impact that considering multiple interconnected domains can have on complex tasks such as gene-disease association prediction. RESULTS: We propose a novel approach to predict gene-disease associations using rich semantic representations based on knowledge graph embeddings over multiple ontologies linked by logical definitions and compound ontology mappings. The experiments showed that considering richer knowledge graphs significantly improves gene-disease prediction and that different knowledge graph embeddings methods benefit more from distinct types of semantic richness. CONCLUSIONS: This work demonstrated the potential for knowledge graph embeddings across multiple and interconnected biomedical ontologies to support gene-disease prediction. It also paved the way for considering other ontologies or tackling other tasks where multiple perspectives over the data can be beneficial. All software and data are freely available.

The Immunopeptidome from a Genomic Perspective: Establishing the Noncanonical Landscape of MHC Class I-Associated Peptides.

Tumor antigens can emerge through multiple mechanisms, including translation of noncoding genomic regions. This noncanonical category of tumor antigens has recently gained attention; however, our understanding of how they recur within and between cancer types is still in its infancy. Therefore, we developed a proteogenomic pipeline based on deep learning de novo mass spectrometry (MS) to enable the discovery of noncanonical MHC class I-associated peptides (ncMAP) from noncoding regions. Considering that the emergence of tumor antigens can also involve posttranslational modifications (PTM), we included an open search component in our pipeline. Leveraging the wealth of MS-based immunopeptidomics, we analyzed data from 26 MHC class I immunopeptidomic studies across 11 different cancer types. We validated the de novo identified ncMAPs, along with the most abundant PTMs, using spectral matching and controlled their FDR to 1%. The noncanonical presentation appeared to be 5 times enriched for the A03 HLA supertype, with a projected population coverage of 55%. The data reveal an atlas of 8,601 ncMAPs with varying levels of cancer selectivity and suggest 17 cancer-selective ncMAPs as attractive therapeutic targets according to a stringent cutoff. In summary, the combination of the open-source pipeline and the atlas of ncMAPs reported herein could facilitate the identification and screening of ncMAPs as targets for T-cell therapies or vaccine development.

Knowledge Graph Embeddings for ICU readmission prediction.

BACKGROUND: Intensive Care Unit (ICU) readmissions represent both a health risk for patients,with increased mortality rates and overall health deterioration, and a financial burden for healthcare facilities. As healthcare became more data-driven with the introduction of Electronic Health Records (EHR), machine learning methods have been applied to predict ICU readmission risk. However, these methods disregard the meaning and relationships of data objects and work blindly over clinical data without taking into account scientific knowledge and context. Ontologies and Knowledge Graphs can help bridge this gap between data and scientific context, as they are computational artefacts that represent the entities of a domain and their relationships to each other in a formalized way. METHODS AND RESULTS: We have developed an approach that enriches EHR data with semantic annotations to ontologies to build a Knowledge Graph. A patient's ICU stay is represented by Knowledge Graph embeddings in a contextualized manner, which are used by machine learning models to predict 30-days ICU readmissions. This approach is based on several contributions: (1) an enrichment of the MIMIC-III dataset with patient-oriented annotations to various biomedical ontologies; (2) a Knowledge Graph that defines patient data with biomedical ontologies; (3) a predictive model of ICU readmission risk that uses Knowledge Graph embeddings; (4) a variant of the predictive model that targets different time points during an ICU stay. Our predictive approaches outperformed both a baseline and state-of-the-art works achieving a mean Area Under the Receiver Operating Characteristic Curve of 0.827 and an Area Under the Precision-Recall Curve of 0.691. The application of this novel approach to help clinicians decide whether a patient can be discharged has the potential to prevent the readmission of [Formula: see text] of Intensive Care Unit patients, without unnecessarily prolonging the stay of those who would not require it. CONCLUSION: The coupling of semantic annotation and Knowledge Graph embeddings affords two clear advantages: they consider scientific context and they are able to build representations of EHR information of different types in a common format. This work demonstrates the potential for impact that integrating ontologies and Knowledge Graphs into clinical machine learning applications can have.

Ontologies and Knowledge Graphs in Oncology Research.

The complexity of cancer research stems from leaning on several biomedical disciplines for relevant sources of data, many of which are complex in their own right. A holistic view of cancer-which is critical for precision medicine approaches-hinges on integrating a variety of heterogeneous data sources under a cohesive knowledge model, a role which biomedical ontologies can fill. This study reviews the application of ontologies and knowledge graphs in cancer research. In total, our review encompasses 141 published works, which we categorized under 14 hierarchical categories according to their usage of ontologies and knowledge graphs. We also review the most commonly used ontologies and newly developed ones. Our review highlights the growing traction of ontologies in biomedical research in general, and cancer research in particular. Ontologies enable data accessibility, interoperability and integration, support data analysis, facilitate data interpretation and data mining, and more recently, with the emergence of the knowledge graph paradigm, support the application of Artificial Intelligence methods to unlock new knowledge from a holistic view of the available large volumes of heterogeneous data.

A Collection of Benchmark Data Sets for Knowledge Graph-based Similarity in the Biomedical Domain.

The ability to compare entities within a knowledge graph is a cornerstone technique for several applications, ranging from the integration of heterogeneous data to machine learning. It is of particular importance in the biomedical domain, where semantic similarity can be applied to the prediction of protein-protein interactions, associations between diseases and genes, cellular localization of proteins, among others. In recent years, several knowledge graph-based semantic similarity measures have been developed, but building a gold standard data set to support their evaluation is non-trivial. We present a collection of 21 benchmark data sets that aim at circumventing the difficulties in building benchmarks for large biomedical knowledge graphs by exploiting proxies for biomedical entity similarity. These data sets include data from two successful biomedical ontologies, Gene Ontology and Human Phenotype Ontology, and explore proxy similarities calculated based on protein sequence similarity, protein family similarity, protein-protein interactions and phenotype-based gene similarity. Data sets have varying sizes and cover four different species at different levels of annotation completion. For each data set, we also provide semantic similarity computations with state-of-the-art representative measures. Database URL: https://github.com/liseda-lab/kgsim-benchmark.

Evolving knowledge graph similarity for supervised learning in complex biomedical domains.

BACKGROUND: In recent years, biomedical ontologies have become important for describing existing biological knowledge in the form of knowledge graphs. Data mining approaches that work with knowledge graphs have been proposed, but they are based on vector representations that do not capture the full underlying semantics. An alternative is to use machine learning approaches that explore semantic similarity. However, since ontologies can model multiple perspectives, semantic similarity computations for a given learning task need to be fine-tuned to account for this. Obtaining the best combination of semantic similarity aspects for each learning task is not trivial and typically depends on expert knowledge. RESULTS: We have developed a novel approach, evoKGsim, that applies Genetic Programming over a set of semantic similarity features, each based on a semantic aspect of the data, to obtain the best combination for a given supervised learning task. The approach was evaluated on several benchmark datasets for protein-protein interaction prediction using the Gene Ontology as the knowledge graph to support semantic similarity, and it outperformed competing strategies, including manually selected combinations of semantic aspects emulating expert knowledge. evoKGsim was also able to learn species-agnostic models with different combinations of species for training and testing, effectively addressing the limitations of predicting protein-protein interactions for species with fewer known interactions. CONCLUSIONS: evoKGsim can overcome one of the limitations in knowledge graph-based semantic similarity applications: the need to expertly select which aspects should be taken into account for a given application. Applying this methodology to protein-protein interaction prediction proved successful, paving the way to broader applications.

Improving the interoperability of biomedical ontologies with compound alignments.

BACKGROUND: Ontologies are commonly used to annotate and help process life sciences data. Although their original goal is to facilitate integration and interoperability among heterogeneous data sources, when these sources are annotated with distinct ontologies, bridging this gap can be challenging. In the last decade, ontology matching systems have been evolving and are now capable of producing high-quality mappings for life sciences ontologies, usually limited to the equivalence between two ontologies. However, life sciences research is becoming increasingly transdisciplinary and integrative, fostering the need to develop matching strategies that are able to handle multiple ontologies and more complex relations between their concepts. RESULTS: We have developed ontology matching algorithms that are able to find compound mappings between multiple biomedical ontologies, in the form of ternary mappings, finding for instance that "aortic valve stenosis"(HP:0001650) is equivalent to the intersection between "aortic valve"(FMA:7236) and "constricted" (PATO:0001847). The algorithms take advantage of search space filtering based on partial mappings between ontology pairs, to be able to handle the increased computational demands. The evaluation of the algorithms has shown that they are able to produce meaningful results, with precision in the range of 60-92% for new mappings. The algorithms were also applied to the potential extension of logical definitions of the OBO and the matching of several plant-related ontologies. CONCLUSIONS: This work is a first step towards finding more complex relations between multiple ontologies. The evaluation shows that the results produced are significant and that the algorithms could satisfy specific integration needs.

Tackling the challenges of matching biomedical ontologies.

BACKGROUND: Biomedical ontologies pose several challenges to ontology matching due both to the complexity of the biomedical domain and to the characteristics of the ontologies themselves. The biomedical tracks in the Ontology Matching Evaluation Initiative (OAEI) have spurred the development of matching systems able to tackle these challenges, and benchmarked their general performance. In this study, we dissect the strategies employed by matching systems to tackle the challenges of matching biomedical ontologies and gauge the impact of the challenges themselves on matching performance, using the AgreementMakerLight (AML) system as the platform for this study. RESULTS: We demonstrate that the linear complexity of the hash-based searching strategy implemented by most state-of-the-art ontology matching systems is essential for matching large biomedical ontologies efficiently. We show that accounting for all lexical annotations (e.g., labels and synonyms) in biomedical ontologies leads to a substantial improvement in F-measure over using only the primary name, and that accounting for the reliability of different types of annotations generally also leads to a marked improvement. Finally, we show that cross-references are a reliable source of information and that, when using biomedical ontologies as background knowledge, it is generally more reliable to use them as mediators than to perform lexical expansion. CONCLUSIONS: We anticipate that translating traditional matching algorithms to the hash-based searching paradigm will be a critical direction for the future development of the field. Improving the evaluation carried out in the biomedical tracks of the OAEI will also be important, as without proper reference alignments there is only so much that can be ascertained about matching systems or strategies. Nevertheless, it is clear that, to tackle the various challenges posed by biomedical ontologies, ontology matching systems must be able to efficiently combine multiple strategies into a mature matching approach.

Semantic Similarity in the Gene Ontology.

Gene Ontology-based semantic similarity (SS) allows the comparison of GO terms or entities annotated with GO terms, by leveraging on the ontology structure and properties and on annotation corpora. In the last decade the number and diversity of SS measures based on GO has grown considerably, and their application ranges from functional coherence evaluation, protein interaction prediction, and disease gene prioritization.Understanding how SS measures work, what issues can affect their performance and how they compare to each other in different evaluation settings is crucial to gain a comprehensive view of this area and choose the most appropriate approaches for a given application.In this chapter, we provide a guide to understanding and selecting SS measures for biomedical researchers. We present a straightforward categorization of SS measures and describe the main strategies they employ. We discuss the intrinsic and external issues that affect their performance, and how these can be addressed. We summarize comparative assessment studies, highlighting the top measures in different settings, and compare different implementation strategies and their use. Finally, we discuss some of the extant challenges and opportunities, namely the increased semantic complexity of GO and the need for fast and efficient computation, pointing the way towards the future generation of SS measures.

Ontology Alignment Repair through Modularization and Confidence-Based Heuristics.

Ontology Matching aims at identifying a set of semantic correspondences, called an alignment, between related ontologies. In recent years, there has been a growing interest in efficient and effective matching methods for large ontologies. However, alignments produced for large ontologies are often logically incoherent. It was only recently that the use of repair techniques to improve the coherence of ontology alignments began to be explored. This paper presents a novel modularization technique for ontology alignment repair which extracts fragments of the input ontologies that only contain the necessary classes and relations to resolve all detectable incoherences. The paper presents also an alignment repair algorithm that uses a global repair strategy to minimize both the degree of incoherence and the number of mappings removed from the alignment, while overcoming the scalability problem by employing the proposed modularization technique. Our evaluation shows that our modularization technique produces significantly small fragments of the ontologies and that our repair algorithm produces more complete alignments than other current alignment repair systems, while obtaining an equivalent degree of incoherence. Additionally, we also present a variant of our repair algorithm that makes use of the confidence values of the mappings to improve alignment repair. Our repair algorithm was implemented as part of AgreementMakerLight, a free and open-source ontology matching system.

Automatic background knowledge selection for matching biomedical ontologies.

Ontology matching is a growing field of research that is of critical importance for the semantic web initiative. The use of background knowledge for ontology matching is often a key factor for success, particularly in complex and lexically rich domains such as the life sciences. However, in most ontology matching systems, the background knowledge sources are either predefined by the system or have to be provided by the user. In this paper, we present a novel methodology for automatically selecting background knowledge sources for any given ontologies to match. This methodology measures the usefulness of each background knowledge source by assessing the fraction of classes mapped through it over those mapped directly, which we call the mapping gain. We implemented this methodology in the AgreementMakerLight ontology matching framework, and evaluate it using the benchmark biomedical ontology matching tasks from the Ontology Alignment Evaluation Initiative (OAEI) 2013. In each matching problem, our methodology consistently identified the sources of background knowledge that led to the highest improvements over the baseline alignment (i.e., without background knowledge). Furthermore, our proposed mapping gain parameter is strongly correlated with the F-measure of the produced alignments, thus making it a good estimator for ontology matching techniques based on background knowledge.

The epidemiology ontology: an ontology for the semantic annotation of epidemiological resources.

BACKGROUND: Epidemiology is a data-intensive and multi-disciplinary subject, where data integration, curation and sharing are becoming increasingly relevant, given its global context and time constraints. The semantic annotation of epidemiology resources is a cornerstone to effectively support such activities. Although several ontologies cover some of the subdomains of epidemiology, we identified a lack of semantic resources for epidemiology-specific terms. This paper addresses this need by proposing the Epidemiology Ontology (EPO) and by describing its integration with other related ontologies into a semantic enabled platform for sharing epidemiology resources. RESULTS: The EPO follows the OBO Foundry guidelines and uses the Basic Formal Ontology (BFO) as an upper ontology. The first version of EPO models several epidemiology and demography parameters as well as transmission of infection processes, participants and related procedures. It currently has nearly 200 classes and is designed to support the semantic annotation of epidemiology resources and data integration, as well as information retrieval and knowledge discovery activities. CONCLUSIONS: EPO is under active development and is freely available at https://code.google.com/p/epidemiology-ontology/. We believe that the annotation of epidemiology resources with EPO will help researchers to gain a better understanding of global epidemiological events by enhancing data integration and sharing.

Predicting the extension of biomedical ontologies.

Developing and extending a biomedical ontology is a very demanding task that can never be considered complete given our ever-evolving understanding of the life sciences. Extension in particular can benefit from the automation of some of its steps, thus releasing experts to focus on harder tasks. Here we present a strategy to support the automation of change capturing within ontology extension where the need for new concepts or relations is identified. Our strategy is based on predicting areas of an ontology that will undergo extension in a future version by applying supervised learning over features of previous ontology versions. We used the Gene Ontology as our test bed and obtained encouraging results with average f-measure reaching 0.79 for a subset of biological process terms. Our strategy was also able to outperform state of the art change capturing methods. In addition we have identified several issues concerning prediction of ontology evolution, and have delineated a general framework for ontology extension prediction. Our strategy can be applied to any biomedical ontology with versioning, to help focus either manual or semi-automated extension methods on areas of the ontology that need extension.

Mining GO annotations for improving annotation consistency.

Despite the structure and objectivity provided by the Gene Ontology (GO), the annotation of proteins is a complex task that is subject to errors and inconsistencies. Electronically inferred annotations in particular are widely considered unreliable. However, given that manual curation of all GO annotations is unfeasible, it is imperative to improve the quality of electronically inferred annotations. In this work, we analyze the full GO molecular function annotation of UniProtKB proteins, and discuss some of the issues that affect their quality, focusing particularly on the lack of annotation consistency. Based on our analysis, we estimate that 64% of the UniProtKB proteins are incompletely annotated, and that inconsistent annotations affect 83% of the protein functions and at least 23% of the proteins. Additionally, we present and evaluate a data mining algorithm, based on the association rule learning methodology, for identifying implicit relationships between molecular function terms. The goal of this algorithm is to assist GO curators in updating GO and correcting and preventing inconsistent annotations. Our algorithm predicted 501 relationships with an estimated precision of 94%, whereas the basic association rule learning methodology predicted 12,352 relationships with a precision below 9%.

Chemical Entity Recognition and Resolution to ChEBI.

Chemical entities are ubiquitous through the biomedical literature and the development of text-mining systems that can efficiently identify those entities are required. Due to the lack of available corpora and data resources, the community has focused its efforts in the development of gene and protein named entity recognition systems, but with the release of ChEBI and the availability of an annotated corpus, this task can be addressed. We developed a machine-learning-based method for chemical entity recognition and a lexical-similarity-based method for chemical entity resolution and compared them with Whatizit, a popular-dictionary-based method. Our methods outperformed the dictionary-based method in all tasks, yielding an improvement in F-measure of 20% for the entity recognition task, 2-5% for the entity-resolution task, and 15% for combined entity recognition and resolution tasks.

Application of gene ontology to gene identification.

Candidate gene identification deals with associating genes to underlying biological phenomena, such as diseases and specific disorders. It has been shown that classes of diseases with similar phenotypes are caused by functionally related genes. Currently, a fair amount of knowledge about the functional characterization can be found across several public databases; however, functional descriptors can be ambiguous, domain specific, and context dependent. In order to cope with these issues, the Gene Ontology (GO) project developed a bio-ontology of broad scope and wide applicability. Thus, the structured and controlled vocabulary of terms provided by the GO project describing the biological roles of gene products can be very helpful in candidate gene identification approaches. The method presented here uses GO annotation data in order to identify the most meaningful functional aspects occurring in a given set of related gene products. The method measures this meaningfulness by calculating an e-value based on the frequency of annotation of each GO term in the set of gene products versus the total frequency of annotation. Then after selecting a GO term related to the underlying biological phenomena being studied, the method uses semantic similarity to rank the given gene products that are annotated to the term. This enables the user to further narrow down the list of gene products and identify those that are more likely of interest.

Rescue of F508del-CFTR by RXR motif inactivation triggers proteome modulation associated with the unfolded protein response.

F508del-CFTR, the most common mutation of the cystic fibrosis transmembrane conductance regulator (CFTR) protein, disrupts intracellular trafficking leading to cystic fibrosis (CF). The trafficking defect of F508del-CFTR can be rescued by simultaneous inactivation of its four RXR motifs (4RK). Proteins involved in the F508del-CFTR trafficking defect and/or rescue are therefore potential CF therapeutic targets. We sought to identify these proteins by investigating differential proteome modulation in BHK cells over-expressing wt-CFTR, F508del-CFTR or the revertant F508del/4RK-CFTR. By 2-dimensional electrophoresis-based proteomics and western blot approaches we demonstrated that over-expression of F508del/4RK-CFTR modulates the expression of a large number of proteins, many of which are reported interactors of CFTR and/or 14-3-3 with potential roles in CFTR trafficking. GRP78/BiP, a marker of ER stress and unfolded protein response (UPR), is up-regulated in cells over-expressing either F508del-CFTR or F598del/4RK-CFTR. However, over-expression of F508del/4RK-CFTR induces the up-regulation of many other UPR-associated proteins (e.g. GRP94, PDI, GRP75/mortalin) and, interestingly, the down-regulation of proteasome components associated with CFTR degradation, such as the proteasome activator PA28 (PSME2) and COP9 signalosome (COPS5/CSN5). Moreover, the F508del-CFTR-induced proteostasis imbalance, which involves some heat shock chaperones (e.g. HSP72/Hpa2), ER-EF-hand Ca(2+)-binding proteins (calumenin) and the proteasome activator PA28 (PSME2), tends to be 'restored', i.e., in BHK cells over-expressing F508del/4RK-CFTR those proteins tend to have expression levels similar to the wild-type ones. These findings indicate that a particular cellular environment orchestrated by the UPR contributes to and/or is compatible with F508del/4RK-CFTR rescue.

Semantic similarity in biomedical ontologies.

In recent years, ontologies have become a mainstream topic in biomedical research. When biological entities are described using a common schema, such as an ontology, they can be compared by means of their annotations. This type of comparison is called semantic similarity, since it assesses the degree of relatedness between two entities by the similarity in meaning of their annotations. The application of semantic similarity to biomedical ontologies is recent; nevertheless, several studies have been published in the last few years describing and evaluating diverse approaches. Semantic similarity has become a valuable tool for validating the results drawn from biomedical studies such as gene clustering, gene expression data analysis, prediction and validation of molecular interactions, and disease gene prioritization. We review semantic similarity measures applied to biomedical ontologies and propose their classification according to the strategies they employ: node-based versus edge-based and pairwise versus groupwise. We also present comparative assessment studies and discuss the implications of their results. We survey the existing implementations of semantic similarity measures, and we describe examples of applications to biomedical research. This will clarify how biomedical researchers can benefit from semantic similarity measures and help them choose the approach most suitable for their studies.Biomedical ontologies are evolving toward increased coverage, formality, and integration, and their use for annotation is increasingly becoming a focus of both effort by biomedical experts and application of automated annotation procedures to create corpora of higher quality and completeness than are currently available. Given that semantic similarity measures are directly dependent on these evolutions, we can expect to see them gaining more relevance and even becoming as essential as sequence similarity is today in biomedical research.

Dynamic modularity in protein interaction networks predicts breast cancer outcome.

Changes in the biochemical wiring of oncogenic cells drives phenotypic transformations that directly affect disease outcome. Here we examine the dynamic structure of the human protein interaction network (interactome) to determine whether changes in the organization of the interactome can be used to predict patient outcome. An analysis of hub proteins identified intermodular hub proteins that are co-expressed with their interacting partners in a tissue-restricted manner and intramodular hub proteins that are co-expressed with their interacting partners in all or most tissues. Substantial differences in biochemical structure were observed between the two types of hubs. Signaling domains were found more often in intermodular hub proteins, which were also more frequently associated with oncogenesis. Analysis of two breast cancer patient cohorts revealed that altered modularity of the human interactome may be useful as an indicator of breast cancer prognosis.