To accelerate the time- and labor-intensive processes of drug discovery and repurposing, it is increasingly common to mine knowledge sources for connections between diseases and the drugs that can treat them. In this article we address the scalability challenge in the connection mining, by introducing algorithms that can be used to find plausible mechanistic connections between drugs and the potentially associated diseases in biomedical knowledge graphs. These connections are then presented to biomedical experts as candidate hypotheses for further studies of whether the drugs can be repurposed to treat the diseases. One challenge that has to be addressed in this effort is the processing of promiscuous knowledge-graph nodes, that is, nodes associated with numerous relationships that may not be unique or indicative of the node properties. As it turns out, the multiplicity of relationships involving promiscuous graph nodes may prevent the aforementioned path-finding algorithms from aiding in drug repurposing. To address the promiscuous-node challenge, we introduce promiscuity scores for nodes and paths in knowledge graphs, and incorporate the scores in the proposed path-finding algorithms. We report experimental results that indicate that paths with low-promiscuity scores could be meaningful and of interest to biomedical experts in drug repurposing.

In deep learning for drug discovery, molecular representations are often based on sequences, known as SMILES, which allow for straightforward implementation of natural language processing methodologies, one being the sequence-to-sequence autoencoder. However, we observe that training an autoencoder solely on SMILES is insufficient to learn molecular representations that are semantically meaningful, where semantics are specified by the structural (graph-to-graph) similarities between molecules. We demonstrate by example that SMILES-based autoencoders may map structurally similar molecules to distant codes, resulting in an incoherent latent space that does not necessarily respect the semantic similarities between molecules. To address this shortcoming we propose Semantically-Aware Latent Space Autoencoder (SALSA) for molecular representations: a SMILES-based transformer autoencoder modified with a contrastive task aimed at learning graph-to-graph similarities between molecules. To accomplish this, we develop a novel dataset comprised of sets of structurally similar molecules and opt for a supervised contrastive loss that is able to incorporate full sets of positive samples. We evaluate semantic awareness of SALSA representations by comparing to its ablated counterparts, and show empirically that SALSA learns representations that maintain 1) structural awareness, 2) physicochemical awareness, 3) biological awareness, and 4) semantic continuity.

In the context of using big data to improve healthcare and life sciences and with a specific focus on drug discovery, we consider the problem of formulating biomedical mechanism-of-action (MOA) hypotheses that can explain how specific drugs treat specific diseases. Our aim is to enable scalable mining and interpretation of MOA hypotheses enabling drug discovery and repurposing on large-scale biomedical knowledge graphs (KGs).The approach that we introduce to address this problem centers on expert-guided generation of candidate MOA hypotheses in the form of regular-path KG patterns between the KG nodes for the entities of interest, such as drugs and diseases. We call those patterns that represent promising candidate MOAs functional path patterns (FPPs), and call the proposed approach FPP-Hunter. The results of a drug-disease case study that we have conducted with the biomedical KG ROBOKOP suggest that the proposed approach has the potential to address scalability challenges in forming promising MOA hypotheses using large-scale KGs, in drug repurposing and potentially beyond.

Knowledge graphs (KGs) represent the connections and relationships between real-world entities. We propose a link prediction framework on KGs named Enrichment-Driven GrAph Reasoner (EDGAR) that infers new edges by mining entity-local rules. This approach is based on enrichment analysis, a well-established statistical method used to calculate mechanisms common to a set of differentially expressed genes. EDGAR’s inference results are inherently explainable and rankable, equipped with p-values for statistical significance of each enrichment-based rule. We demonstrate its effectiveness on a large-scale biomedical KG, ROBOKOP, focusing on drug repurposing for Alzheimer disease (AD) as a case study. Initially, we extracted 14 known drugs from the KG and identified 20 contextual biomarkers through enrichment analysis, shedding light on functional pathways relevant to the shared efficacy of drugs for AD. Subsequently, using the top 1,000 enrichment results, our enrichment-driven system identified 1,246 additional drug candidates for AD treatment. We validated the top 10 candidates using medical literature evidence. EDGAR is deployed within ROBOKOP, along with a web user interface. This is the first work to use enrichment analysis for either large graph completion or drug repurposing.

SUMMARY: Knowledge graphs are being increasingly used in biomedical research to link large amounts of heterogenous data and facilitate reasoning across diverse knowledge sources. Wider adoption and exploration of knowledge graphs in the biomedical research community is limited by requirements to understand the underlying graph structure in terms of entity types and relationships, represented as nodes and edges, respectively, and learn specialized query languages for graph mining and exploration. We have developed a user-friendly interface dubbed ExEmPLAR (Extracting, Exploring, and Embedding Pathways Leading to Actionable Research) to aid reasoning over biomedical knowledge graphs and assist with data-driven research and hypothesis generation. We explain the key functionalities of ExEmPLAR and demonstrate its use with a case study considering the relationship of Trypanosoma cruzi, the etiological agent of Chagas disease, to frequently associated cardiovascular conditions. AVAILABILITY AND IMPLEMENTATION: ExEmPLAR is freely accessible at https://www.exemplar.mml.unc.edu/. For code and instructions for the using the application, see: https://github.com/beasleyjonm/AOP-COP-Path-Extractor.

Knowledge graphs have been growing in popularity in recent years, due to their successful use in many data-analytics and knowledge-discovery tasks. As the sizes of domain graphs continue to grow, they can become too large to be processed efficiently by downstream applications. It is thus indispensable in many cases to aim to address the efficiency issue by reducing knowledge-graph sizes via summarization. At the same time, to provide quality results, that is, to be effective, on the graph summaries, many downstream analytics and knowledge-discovery tasks require the summaries to retain as much information from the original graph as possible. It turns out that state-of-the-art data-reduction approaches may generate task-specific or lossy summaries, which may present challenges for the effectiveness of downstream analytics in some cases. As a result, it is important to study trade-offs between the performance and accuracy of graph summarizations for analytics and knowledge discovery.In this paper, in the scope of knowledge-graph size reduction via summarization, we consider the trade-off between the performance and accuracy of one type of knowledge-graph analytics, inference-rule mining, on the summarized graphs as compared with the original graphs. Toward addressing the tradeoff challenge, we introduce a domain- and task-independent knowledge-graph summarization approach called GAME for generating reduced abstract knowledge graphs from a given graph. Our preliminary experimental results suggest that our GAME approach can preserve as much information from the given (original) graph as possible for the purpose of rule mining, while potentially significantly improving the rule-mining efficiency. We anticipate that the knowledge-graph reduction provided by the proposed GAME approach can enable efficient, yet effective, results for diverse applications and use cases.

Recently, knowledge graphs have seen a significant increase in popularity in a wide variety of domains, as they provide a basis for many data-analytics and knowledge-discovery approaches. At the same time, many knowledge graphs are not immediately usable due to their format, unreadable or missing data, and inaccessibility. These issues present barriers to the exploration and use of knowledge graphs for big data analytics and knowledge discovery. In this paper we present a workflow for domain- and task-sensitive curation of large-scale knowledge graphs, and detail our experience with implementing this workflow with the biomedical knowledge graph called Drug Repurposing Knowledge Graph (DRKG). The workflow aims to address usability-related issues of real-life knowledge graphs, by performing data setup and curation that align with the needs of specific tasks and domains. Recognizing that domain experts and anticipated users of a knowledge graph provide invaluable expertise regarding the desired graph format, the proposed workflow involves them as humans-in-the-loop. We present the processes required to execute the workflow, detail our experience in the biomedical domain with the use case of DRKG, and discuss the challenges and lessons learned throughout the experience. We anticipate that the proposed workflow and experiences will be applicable to other domains, and that our workflow will enable and encourage exploration and wider use of large-scale knowledge graphs, thereby improving big data analytics.

Knowledge-graph (KG) embeddings have emerged as a promise in addressing challenges faced by modern biomedical research, including the growing gap between therapeutic needs and available treatments. The popularity of KG embeddings in graph analytics is on the rise, due at least partially to the presumed semanticity of the learned embeddings. Unfortunately, the ability of a node neighborhood picked up by an embedding to capture the node's semantics may depend on the characteristics of the data. One of the reasons for this problem is that KG nodes can be promiscuous, that is, associated with a number of different relationships that are not unique or indicative of the properties of the nodes.

Drug discovery is an important process used by biomedical experts to identify potential treatments for diseases. In its traditional form, the process requires significant expert time and manual effort. By encoding a wealth of information about relationships between drugs and diseases, modern large-scale biomedical knowledge graphs provide excellent opportunities to accelerate drug discovery, by automating aspects of the process. One opportunity is to use explainable fact-checking tools to generate explanations for hypothesized drug-disease treatment relationships in a given knowledge graph, with a reliability score assigned to each explanation. The explanations and their scores can then be used by experts to determine which drug-disease pairs to consider for clinical trials.In our collaboration with a biomedical team, we have found that existing explainable fact-checking tools are not necessarily helpful in drug discovery, as their explanation formats and evaluation metrics do not match well the requirements of scientific discovery in the biomedical domain. To address these challenges in using fact-checking tools in drug discovery, we introduce a scalable automated approach for generating explanations that are modeled after existing biomedical concepts and supplemented with data-supported evaluation metrics. Our explanations are based on knowledge-graph patterns, which are readily understood by biomedical experts. Our experimental results suggest that our proposed metrics are accurate and useful on largescale biomedical knowledge graphs, and our explanations are understandable and reasonable to experts doing drug discovery.

SUMMARY: In response to the COVID-19 pandemic, we established COVID-KOP, a new knowledgebase integrating the existing Reasoning Over Biomedical Objects linked in Knowledge Oriented Pathways (ROBOKOP) biomedical knowledge graph with information from recent biomedical literature on COVID-19 annotated in the CORD-19 collection. COVID-KOP can be used effectively to generate new hypotheses concerning repurposing of known drugs and clinical drug candidates against COVID-19 by establishing respective confirmatory pathways of drug action. AVAILABILITY AND IMPLEMENTATION: COVID-KOP is freely accessible at https://covidkop.renci.org/. For code and instructions for the original ROBOKOP, see: https://github.com/NCATS-Gamma/robokop.

A continuing challenge in modern medicine is the identification of safer and more efficacious drugs. Precision therapeutics, which have one molecular target, have been long promised to be safer and more effective than traditional therapies. This approach has proven to be challenging for multiple reasons including lack of efficacy, rapidly acquired drug resistance, and narrow patient eligibility criteria. An alternative approach is the development of drugs that address the overall disease network by targeting multiple biological targets ('polypharmacology'). Rational development of these molecules will require improved methods for predicting single chemical structures that target multiple drug targets. To address this need, we developed the Multi-Targeting Drug DREAM Challenge, in which we challenged participants to predict single chemical entities that target pro-targets but avoid anti-targets for two unrelated diseases: RET-based tumors and a common form of inherited Tauopathy. Here, we report the results of this DREAM Challenge and the development of two neural network-based machine learning approaches that were applied to the challenge of rational polypharmacology. Together, these platforms provide a potentially useful first step towards developing lead therapeutic compounds that address disease complexity through rational polypharmacology.

MOTIVATION: Non-ribosomal peptide synthetases (NRPSs) are modular enzymatic machines that catalyze the ribosome-independent production of structurally complex small peptides, many of which have important clinical applications as antibiotics, antifungals and anti-cancer agents. Several groups have tried to expand natural product diversity by intermixing different NRPS modules to create synthetic peptides. This approach has not been as successful as anticipated, suggesting that these modules are not fully interchangeable. RESULTS: We explored whether Inter-Modular Linkers (IMLs) impact the ability of NRPS modules to communicate during the synthesis of NRPs. We developed a parser to extract 39 804 IMLs from both well annotated and putative NRPS biosynthetic gene clusters from 39 232 bacterial genomes and established the first IMLs database. We analyzed these IMLs and identified a striking relationship between IMLs and the amino acid substrates of their adjacent modules. More than 92% of the identified IMLs connect modules that activate a particular pair of substrates, suggesting that significant specificity is embedded within these sequences. We therefore propose that incorporating the correct IML is critical when attempting combinatorial biosynthesis of novel NRPS. AVAILABILITY AND IMPLEMENTATION: The IMLs database as well as the NRPS-Parser have been made available on the web at https://nrps-linker.unc.edu. The entire source code of the project is hosted in GitHub repository (https://github.com/SWFarag/nrps-linker). SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

SUMMARY: Knowledge graphs (KGs) are quickly becoming a common-place tool for storing relationships between entities from which higher-level reasoning can be conducted. KGs are typically stored in a graph-database format, and graph-database queries can be used to answer questions of interest that have been posed by users such as biomedical researchers. For simple queries, the inclusion of direct connections in the KG and the storage and analysis of query results are straightforward; however, for complex queries, these capabilities become exponentially more challenging with each increase in complexity of the query. For instance, one relatively complex query can yield a KG with hundreds of thousands of query results. Thus, the ability to efficiently query, store, rank and explore sub-graphs of a complex KG represents a major challenge to any effort designed to exploit the use of KGs for applications in biomedical research and other domains. We present Reasoning Over Biomedical Objects linked in Knowledge Oriented Pathways as an abstraction layer and user interface to more easily query KGs and store, rank and explore query results. AVAILABILITY AND IMPLEMENTATION: An instance of the ROBOKOP UI for exploration of the ROBOKOP Knowledge Graph can be found at http://robokop.renci.org. The ROBOKOP Knowledge Graph can be accessed at http://robokopkg.renci.org. Code and instructions for building and deploying ROBOKOP are available under the MIT open software license from https://github.com/NCATS-Gamma/robokop. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

OBJECTIVE: This study aimed to develop a novel, regulatory-compliant approach for openly exposing integrated clinical and environmental exposures data: the Integrated Clinical and Environmental Exposures Service (ICEES). MATERIALS AND METHODS: The driving clinical use case for research and development of ICEES was asthma, which is a common disease influenced by hundreds of genes and a plethora of environmental exposures, including exposures to airborne pollutants. We developed a pipeline for integrating clinical data on patients with asthma-like conditions with data on environmental exposures derived from multiple public data sources. The data were integrated at the patient and visit level and used to create de-identified, binned, "integrated feature tables," which were then placed behind an OpenAPI. RESULTS: Our preliminary evaluation results demonstrate a relationship between exposure to high levels of particulate matter ≤2.5 µm in diameter (PM2.5) and the frequency of emergency department or inpatient visits for respiratory issues. For example, 16.73% of patients with average daily exposure to PM2.5 >9.62 µg/m3 experienced 2 or more emergency department or inpatient visits for respiratory issues in year 2010 compared with 7.93% of patients with lower exposures (n = 23 093). DISCUSSION: The results validated our overall approach for openly exposing and sharing integrated clinical and environmental exposures data. We plan to iteratively refine and expand ICEES by including additional years of data, feature variables, and disease cohorts. CONCLUSIONS: We believe that ICEES will serve as a regulatory-compliant model and approach for promoting open access to and sharing of integrated clinical and environmental exposures data.

OBJECTIVE: Quantitative Structure-Activity Relationship (QSAR) models can predict adverse drug reactions (ADRs), and thus provide early warnings of potential hazards. Timely identification of potential safety concerns could protect patients and aid early diagnosis of ADRs among the exposed. Our objective was to determine whether global spontaneous reporting patterns might allow chemical substructures associated with Stevens-Johnson Syndrome (SJS) to be identified and utilized for ADR prediction by QSAR models. MATERIALS AND METHODS: Using a reference set of 364 drugs having positive or negative reporting correlations with SJS in the VigiBase global repository of individual case safety reports (Uppsala Monitoring Center, Uppsala, Sweden), chemical descriptors were computed from drug molecular structures. Random Forest and Support Vector Machines methods were used to develop QSAR models, which were validated by external 5-fold cross validation. Models were employed for virtual screening of DrugBank to predict SJS actives and inactives, which were corroborated using knowledge bases like VigiBase, ChemoText, and MicroMedex (Truven Health Analytics Inc, Ann Arbor, Michigan). RESULTS: We developed QSAR models that could accurately predict if drugs were associated with SJS (area under the curve of 75%-81%). Our 10 most active and inactive predictions were substantiated by SJS reports (or lack thereof) in the literature. DISCUSSION: Interpretation of QSAR models in terms of significant chemical descriptors suggested novel SJS structural alerts. CONCLUSIONS: We have demonstrated that QSAR models can accurately identify SJS active and inactive drugs. Requiring chemical structures only, QSAR models provide effective computational means to flag potentially harmful drugs for subsequent targeted surveillance and pharmacoepidemiologic investigations.

SUMMARY: We report on the development of the high-throughput screening (HTS) Navigator software to analyze and visualize the results of HTS of chemical libraries. The HTS Navigator processes output files from different plate readers' formats, computes the overall HTS matrix, automatically detects hits and has different types of baseline navigation and correction features. The software incorporates advanced cheminformatics capabilities such as chemical structure storage and visualization, fast similarity search and chemical neighborhood analysis for retrieved hits. The software is freely available for academic laboratories. AVAILABILITY AND IMPLEMENTATION: http://fourches.web.unc.edu/

Continued surveillance of post-marketing Adverse Drug Events (ADEs) is considered essential for patient safety, and Electronic Health Records (EHRs) serve as a critical source for identifying relevant information. But effective EHR knowledge discovery and data mining is not trivial because involved data usually have significantly different semantics among each other. Semantic technologies are believed to greatly assist in this regard; unfortunately, semantic technologies and conventional data mining remain largely separate disciplines, and the fusion of these two disciplines is still in its infancy. This position paper explores two semantics-driven frequent data pattern mining algorithms for EHR knowledge discovery, aiming at more effective ADE monitoring in a population. By effectively utilizing human knowledge formally encoded in EHR domain ontologies, our proposed algorithms will enhance the identification of the drug ADE causality out of large amounts of heterogeneous data sets. Through mining a large corpus of representative EHRs at semantic level, we will be able to compile a comprehensive list of ADE endpoints by obtaining critical, but originally hidden and implicit, frequent data patterns. Ultimately, our software to be developed will significantly facilitate effective ADE monitoring and prediction. Moreover, our research is expected to produce broader impacts on the pharmaceutical industry by reducing the R & D cost for new drug discovery and on transforming current pharmacovigilance methods to reduce adverse events and hence improve human health.

MOTIVATION: Advances in the field of cheminformatics have been hindered by a lack of freely available tools. We have created Chembench, a publicly available cheminformatics portal for analyzing experimental chemical structure-activity data. Chembench provides a broad range of tools for data visualization and embeds a rigorous workflow for creating and validating predictive Quantitative Structure-Activity Relationship models and using them for virtual screening of chemical libraries to prioritize the compound selection for drug discovery and/or chemical safety assessment. AVAILABILITY: Freely accessible at: http://chembench.mml.unc.edu CONTACT: [email protected]

We describe a new approach for inferring the functional relationships between nonhomologous protein families by looking at statistical enrichment of alternative function predictions in classification hierarchies such as Gene Ontology (GO) and Structural Classification of Proteins (SCOP). Protein structures are represented by robust graph representations, and the fast frequent subgraph mining algorithm is applied to protein families to generate sets of family-specific packing motifs, i.e., amino acid residue-packing patterns shared by most family members but infrequent in other proteins. The function of a protein is inferred by identifying in it motifs characteristic of a known family. We employ these family-specific motifs to elucidate functional relationships between families in the GO and SCOP hierarchies. Specifically, we postulate that two families are functionally related if one family is statistically enriched by motifs characteristic of another family, i.e., if the number of proteins in a family containing a motif from another family is greater than expected by chance. This function-inference method can help annotate proteins of unknown function, establish functional neighbors of existing families, and help specify alternate functions for known proteins.

We describe a new approach for inferring the functional relationships between non-homologous protein families by looking at statistical enrichment of alternative function predictions in classification hierarchies such as Gene Ontology (GO) and Structural Classification of Proteins (SCOP). Protein structures are represented by robust graphs, and the Fast Frequent Subgraph Mining algorithm is applied to protein families to generate sets of family-specific packing motifs, i.e. amino acid residue packing patterns shared by most family members but infrequent in other proteins. The function of a protein is inferred by identifying in it motifs characteristic of a known family. We employ these family-specific motifs to elucidate functional relationships between families in the GO and SCOP hierarchies. Specifically, we postulate that two families are functionally related if one family is statistically enriched by motifs characteristic of another family, i.e. if the number of proteins in a family containing a motif from another family is greater than expected by chance. This function inference method can help annotate proteins of unknown function, establish functional neighbors of existing families, and help specify alternate functions for known proteins.

Finding recurring residue packing patterns, or spatial motifs, that characterize protein structural families is an important problem in bioinformatics. We apply a novel frequent subgraph mining algorithm to three graph representations of protein three-dimensional (3D) structure. In each protein graph, a vertex represents an amino acid. Vertex-residues are connected by edges using three approaches: first, based on simple distance threshold between contact residues; second using the Delaunay tessellation from computational geometry, and third using the recently developed almost-Delaunay tessellation approach.Applying a frequent subgraph mining algorithm to a set of graphs representing a protein family from the Structural Classification of Proteins (SCOP) database, we typically identify several hundred common subgraphs equivalent to common packing motifs found in the majority of proteins in the family. We also use the counts of motifs extracted from proteins in two different SCOP families as input variables in a binary classification experiment. The resulting models are capable of predicting the protein family association with the accuracy exceeding 90 percent. Our results indicate that graphs based on both almost-Delaunay and Delaunay tessellations are sparser than the contact distance graphs; yet they are robust and efficient for mining protein spatial motif.

MOTIVATION: Most scoring functions used in protein fold recognition employ two-body (pseudo) potential energies. The use of higher-order terms may improve the performance of current algorithms. METHODS: Proteins are represented by the side chain centroids of amino acids. Delaunay tessellation of this representation defines all sets of nearest neighbor quadruplets of amino acids. Four-body contact scoring function (log likelihoods of residue quadruplet compositions) is derived by the analysis of a diverse set of proteins with known structures. A test protein is characterized by the total score calculated as the sum of the individual log likelihoods of composing amino acid quadruplets. RESULTS: The scoring function distinguishes native from partially unfolded or deliberately misfolded structures. It also discriminates between pre- and post-transition state and native structures in the folding simulations trajectory of Chymotrypsin Inhibitor 2 (CI2).

Rapid growth of protein structures database in recent years requires an effective approach for objective comparison and classification of deposited protein structures. We describe a novel method for structure comparison and classification based on the alignment of one-dimensional structure profiles. These profiles are obtained by calculating the OCCO pseudodihedral angles (formed by O-C-C-O atoms of carbonyl groups of consecutive amino acid residues) from protein three-dimensional coordinates. These angle measurements are then converted into a 24 letter alphabet, and the protein structures are represented by sequences of letter from this alphabet. The BioSCAN parallel computer, designed for primary sequence alignment, is used to rapidly align and classify these one-dimensional structure profiles. We have developed and implemented weighted scoring matrix to identify structural classes based on commonly found structural motifs. The results of our experiments are in good agreement with the traditional protein structure classification schemes. One-dimensional structure profiles significantly improve efficiency of structure comparison and classification.