BGBenjamin M. Gyori

Benjamin M. Gyori

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Benjamin M. Gyori has a total of 640 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    MSstatsBioNet: Integrating Statistical Analyses with Prior Knowledge Biomolecular Networks for Quantitative Proteomics and Phosphoproteomics

    by Anthony Wu, Devon Kohler, Pruthvi Prakash Navada, Julia E Robbins, Gabriel E Boyle, Alex Boshart, Klas Karis, Jacques Neefjes, Ana Konvalinka, Jay Sarthy, et al.

    A common outcome of quantitative mass spectrometry-based proteomic and phosphoproteomic experiments is a list of proteins that are differentially abundant between conditions. However, biological interpretation requires evaluation in the context of prior knowledge of biological mechanisms and protein function. One approach to facilitate mechanistic biological interpretation is to integrate such lists with biological network databases, built from manually curated resources and text mining systems. This manuscript automates this process with MSstatsBioNet, a Bioconductor package that integrates MSstats, a family of open-source packages for detecting differentially abundant proteins, and INDRA, a system that extracts biomolecular networks from biomedical literature using text mining and merges those networks with the content of curated knowledge bases. Taking as input a list of differentially abundant proteins from MSstats, MSstatsBioNet retrieves a protein subnetwork from INDRA and overlays experimental fold changes onto the underlying subnetwork. Users can then interact with the network and overlaid data, interrogating primary literature evidence to construct granular mechanistic narratives for iterative hypothesis generation. We demonstrate the utility of this approach with three case studies, two measuring changes in protein abundance and one measuring changes in phosphorylation.

  • Open Access

    A Multimodal Benchmark for Evaluating Cause-of-Death Inference Using Child Health and Mortality Data

    by Junhe Yang, Soumyakanti Pan, Hyun Seung Lim, Yue Chu, Yuting Guo, Nishtha Agarwal, Varun Babbar, Gaurav Rajesh Parikh, Yiqun T. Chen, Chris A. Rees, et al.

    A bs t r act: Accurately attributing causes of death is vital for global health, yet fewer than 5% of deaths in resource-constrained regions are medically certified. To assign causes to these unlabeled deaths at scale, practitioners traditionally rely on verbal autopsy, using supervised statistical models to classify based on structured survey data. However, modern mortality surveillance increasingly collects rich, unstructured multimodal data, such as free-text caregiver narratives and postmortem diagnostics, which traditional supervised statistical models struggle to seamlessly integrate. In this paper, we present a comprehensive, multimodal benchmark for cause-of-death classification using data from the Child Health and Mortality Prevention Surveillance (CHAMPS) network, a unique surveillance platform spanning nine countries across South Asia and Sub-Saharan Africa. Using this dataset, we introduce an evaluation framework designed to rigorously assess diagnostic reasoning, moving beyond traditional metrics that fail to capture complex clinical realities. We demonstrate the utility of this benchmark by evaluating zero-shot large language models against supervised baselines across various data modalities. Our results reveal distinct differences in how these modeling approaches synthesize unstructured medical evidence. This benchmark provide a rigorously defined resource for assessing clinical reasoning in next-generation mortality surveillance.

  • Open Access

    Weakly Supervised Representation Learning for Cross-Ontology Mapping

    by Harshit Soni, Gabriel Nixon, Woodward Galbraith, Benjamin M Gyori

    Formal ontologies capture a domain’s concepts and their relationships, making them essential for principled data integration in biomedicine and other fields. However, since multiple domain ontologies are developed independently, they frequently contain overlapping concepts. As such, automated ontology mapping methods that identify semantically equivalent or related concepts across ontologies are essential for data integration. A primary bottleneck for ontology mapping methods is candidate retrieval, which identifies sets of potentially related concepts. This is the first of three steps in most contemporary ontology mapping systems (such as BERTMap [1]) that typically consist of (i) candidate retrieval, (ii) ranking or scoring of candidates for mapping (often using neural approaches), and (iii) selection of mappings followed by any necessary logical repairs. Since only retrieved candidates are considered for scoring and mapping, retrieval failures can substantially limit achievable recall. To address this bottleneck, we propose a framework for ontology mapping that improves candidate retrieval using weak semantic priors and conservative representation adaptation. Our approach builds on SapBERT [4], an encoder model trained on the UMLS system for linking free text to concepts in biomedical ontologies. Although SapBERT provides strong general representations, its embedding space remains static and may not optimally reflect the semantic geometry required for specific ontology mapping tasks. To adapt representations without introducing supervision leakage, we leverage SeMRA’s (Semantic Mapping Reasoning Assembler) Raw Semantic Mapping Database [3] as a weak semantic prior. SeMRA aggregates cross-resource biomedical mappings at scale, providing high coverage but noisy alignments. Rather than treating these mappings as a gold standard, we interpret them as weak positive signals and fine-tune SapBERT using a positive-only contrastive objective with in-batch negatives (Multiple Negatives Ranking Loss). This formulation encourages semantically related concepts to move closer in the embedding space while implicitly separating unrelated concepts. The resulting embedding-based retrieval method is modular and model-agnostic, replacing existing candidate retrieval methods (e.g., TF-IDF) without modifying downstream ranking or repair components. We used our framework to map the Mammalian Phenotype Ontology (MP) and the Human Phenotype Ontology(HP), two independently curated but biologically related ontologies with substantial lexical and structural differences. This pair serves as a well-established benchmark for biological ontology mapping. The candidate set produced by our method was evaluated using gold standard mappings [5]. Compared to lexical retrieval (Recall@50 ≈ 0.66) [1], embedding-based candidate retrieval improved coverage to ≈ 0.84, and weakly supervised fine-tuning further increased it to ≈ 0.89. When integrated into a BERTMap-style pipeline with fixed candidate sets of size 50, these gains translated into nearly doubling (from 0.3 to 0.65 in MP to HP, 0.3 to 0.4 in HP to MP) alignment accuracy (Recall@1), demonstrating that candidate retrieval rather than scoring, is the primary performance limiter. These findings demonstrate that candidate retrieval is an important bottleneck in the performance of ontology mapping algorithms but can be improved using weak semantic priors combined with positive-only contrastive adaptation. One limitation of our current approach is that prior mappings are treated as binary. Future work will incorporate confidence weights and explicit negative examples to further improve performance. We also aim to construct richer semantic priors by integrating biomedical knowledge graphs and other mapping resources into a unified weak supervision framework. Finally, an integrated mapping system using our retrieval approach will be connected to Biomappings [2] to facilitate expert curation across a larger space of biomedical ontologies.

  • Open Access

    Polypharmacology is an enduring and nearly universal property of kinase inhibitors

    by Caitlin E. Mills, Clemens Hug, Karuna Anna Sajeevan, Nicholas Clark, Chiara Victor, Mirra Chung, Sameer Rawat, Hope D’ERasmo, Mariana Pereira Moraes, Bree Aldridge, et al.

    Despite decades of research, current understanding of the spectrum of targets bound by kinase inhibitors remains incomplete. This complicates mechanism of action studies, drug repurposing, and development of new therapies. Here, we describe kinome-wide profiling of an optimal kinase library (OKL) comprising 192 small molecules selected based on stage of clinical development, chemical diversity, and target coverage. Our results show that polypharmacology is widespread and independent of regulatory approval. The generally understood (“assigned”) targets of approved molecules are not necessarily the most potently inhibited and off-targets include multiple understudied kinases. Moreover, median selectivity has not increased over time We illustrate how an OKL in combination with detailed kinome profiling can be used to identify potential toxicity targets, repurpose anti-inflammatory drugs for neurodegenerative and infectious diseases, and perform chemical genetic studies. Our studies also highlight how much remains to be discovered about the chemistry and biology of one of the largest classes of human therapeutics.

  • Open Access

    Literature-driven extraction and computational prediction of causal statements linking genetic variants to biological processes, pathways and phenotypes

    by Jici Jiang, Predrag Radivojac, Benjamin M. Gyori

    Understanding the mechanistic basis of pathogenic genetic variants requires reconstructing the molecular pathways connecting the variant, via a chain of molecular intermediates, to a disease-causing biological process and phenotype. However, a literature-wide assembly of causal networks connecting variants, molecular pathways, biological processes and phenotypes has not been previously available. To create such a resource, we developed an automated pathway reconstruction approach building on the Integrated Network and Dynamical Reasoning Assembler (INDRA) system which extracts causal mechanistic statements (positive regulation, phosphorylation, complex formation, etc.) by combining structured databases and literature mining. We traversed INDRA statements extracted from publications to identify those describing a genetic variant resulting in a protein point mutation. We then reconstructed directed paths (consisting of one or more linked INDRA statements) connecting this variant to a term representing a biological process, phenotype or disease within the same publication. This resulted in a directed multigraph obtained from 25,862 paths for variants in 2,561 proteins. Each node in this graph corresponds to an ontology-grounded molecular or process term and each edge is explicitly linked to supporting literature evidence, enabling full auditability of inferred mechanisms. To leverage the assembled networks, we trained a classification model to predict likely downstream biological processes or specific disease associations for protein variants. As features to the model, we integrated molecular annotations (including protein sequence features, ClinVar pathogenicity labels, and UniProt domain mappings) in combination with representations from the ESM2 transformer-based protein language model. The performance achieved by this model shows promise for reconstructing causal mechanistic statements associated with function of genetic variants, a framing of the variant effect prediction task that goes significantly beyond simple assessment of pathogenicity. This integrative framework enables the mechanistic interpretation of known variants and prediction of functional relevance for variants lacking prior phenotypic annotation.

  • Open Access

    Revealing Interactions between Microbes, Metabolites, and Dietary Compounds using Genome-scale Analysis

    by Tong Wang, Benjamin Gyori, Scott T. Weiss, Giulia Menichetti, Yang-Yu Liu

    Background: The role of gut microbiome in predicting diet response and developing personalized dietary recommendations has been increasingly recognized. Yet, we still lack comprehensive, genome-based insights into which gut microbes metabolize specific dietary compounds. Results: Here, we leveraged the metabolic networks constructed from well-annotated microbial genomes to characterize the potential interactions between microbes and metabolites, specifically emphasizing the interactions between microbes and dietary compounds. We revealed a substantial, approximately four-fold variation in both the number of metabolites and dietary compounds in the microbial genome-scale metabolic networks across different genera, whereas species within the same genus showed a high metabolic similarity (mean coefficient of variation in microbial network degree CV = 0.023 for metabolites and 0.015 for dietary compounds). We found that the number of species that can utilize a metabolite drastically varies, ranging from 1 to 818 species, with some metabolites being used by a wide range of species (211 out of 1390 metabolites used by more than 95% of species) and others only by a few species (435 metabolites used by less than 5% of species). Leveraging a longitudinal microbiome study, we observed that microbial taxa with similar metabolic capacity tend to have positively correlated abundances, and the gut microbiome’s capacity to process dietary compounds is functionally stable. Finally, we propose a network-based method to identify the dietary compounds that are specific to no more than 10 microbial species, offering a new strategy for combining a dietary compound and its linked microbial species to design synbiotics. Conclusions: Our results quantitatively reveal large-scale variation and redundancy in gut microbial metabolism and identify dietary compounds linked to only a few microbial species. These findings improve understanding of microbe-metabolite interactions and provide a foundation for the rational design of microbiome-based interventions for healthy benefits.

  • Open Access

    SIGNOR 4.0: the 2025 update with focus on phosphorylation data

    by Prisca Lo Surdo, Marta Iannuccelli, Klas Karis, Eleonora Meo, Parnian Omidi, Marta Tosoni, Simone Graziosi, Simona Panni, Maria Teresa Fiorillo, Luana Licata, et al.

    The SIGnaling Network Open Resource (SIGNOR 4.0, https://signor.uniroma2.it) is a database of manually curated causal interactions between biological entities. These signaling events are annotated along with their effect—denoting the activation or inactivation of a target entity—and mechanism through which it is mediated, such as phosphorylation, binding, or transcriptional regulation. The data is freely accessible and can be explored as customizable signaling networks, allowing users to adapt them for different modeling purposes. In our latest update (version 4.0), we improved our curation interface to include additional data-validation tools, integrated text-mining-assisted curation, increased our curation content, and expanded the scope of our curation efforts, with a particular emphasis on phosphorylation data. Furthermore, we developed a subdomain of SIGNOR, PhosphoSIGNOR, a dedicated user interface designed to enable targeted access to phosphorylation-specific information and network visualization. This expanded dataset allows for a more comprehensive mapping of signaling alterations and their associations with dysregulated cellular processes. The platform enables users to dynamically query phosphosite-specific data, examine context-dependent modifications, and integrate findings with known regulatory mechanisms. The PhosphoSIGNOR section of SIGNOR serves as an essential resource for cancer systems biology, offering an intuitive interface for hypothesis generation and mechanistic insights.


  • Open Access

    Assembly and reasoning over semantic mappings at scale for biomedical data integration

    by Charles Tapley Hoyt, Klas Karis, Benjamin M Gyori

    Motivation: Hundreds of resources assign identifiers to biomedical concepts including genes, small molecules, biological processes, diseases, and cell types. Often, these resources overlap by assigning identifiers to the same or related concepts. This creates a data interoperability bottleneck, as integrating data sets and knowledge bases that use identifiers for the same concepts from different resources requires such identifiers to be mapped to each other. However, available mappings are incomplete and fragmented across individual resources, motivating their large-scale integration. Results: We developed SeMRA, a software tool that integrates mappings from multiple sources into a graph data structure. Using graph algorithms, it infers missing mappings implied by available ones while keeping track of provenance and confidence. This allows connecting identifier spaces between which direct mapping was previously not possible. SeMRA implements a customizable workflow that takes a declarative specification as input describing sources to integrate with additional configuration parameters. We used SeMRA to produce the SeMRA Raw Mappings Database, an aggregation of 43.4 million mappings from 127 sources that jointly cover identifiers from 445 ontologies and databases. We also describe benchmarks on specific use cases such as integrating mappings between resources cataloging diseases and cell types. Availability and implementation: The code is available under the MIT license at https://github.com/biopragmatics/semra. The SeMRA Raw Mappings Database assembled by SeMRA is available at https://doi.org/10.5281/zenodo.11082038.


  • Open Access

    Assembly and reasoning over semantic mappings at scale for biomedical data integration

    by Charles Tapley Hoyt, Klas Karis, Benjamin M. Gyori

    Motivation Hundreds of resources assign identifiers to biomedical concepts including genes, small molecules, biological processes, diseases, and cell types. Often, these resources overlap by assigning identifiers to the same or related concepts. This creates a data interoperability bottleneck, as integrating data sets and knowledge bases that use identifiers for the same concepts from different resources require such identifiers to be mapped to each other. However, available mappings are incomplete and fragmented across individual resources, motivating their large-scale integration. Results We developed SeMRA, a software tool that integrates mappings from multiple sources into a graph data structure. Using graph algorithms, it infers missing mappings implied by available ones while keeping track of provenance and confidence. This allows connecting identifier spaces between which direct mapping was previously not possible. SeMRA is customizable and takes a declarative specification as input describing sources to integrate with additional configuration parameters. We make available an aggregated mappings resource produced by SeMRA consisting of 43.4 million mappings from 127 sources that jointly cover identifiers from 445 ontologies and databases. We also describe benchmarks on specific use cases such as integrating mappings between resources cataloging diseases or cell types. Availability The code is available under the MIT license at https://github.com/biopragmatics/semra. The mappings database assembled by SeMRA is available at https://zenodo.org/records/15208251. Contact cthoyt{at}gmail.com and b.gyori{at}northeastern.edu

  • Beyond protein lists: AI-assisted interpretation of proteomic investigations in the context of evolving scientific knowledge

    by Benjamin M. Gyori, Olga Vitek

    Mass spectrometry-based proteomics provides broad and quantitative detection of the proteome, but its results are mostly presented as protein lists. Artificial intelligence approaches will exploit prior knowledge from literature and harmonize fragmented datasets to enable mechanistic and functional interpretation of proteomics experiments.