Biology subjects

McDonagh, E. M.

Publications and source records attributed to McDonagh, E. M..

6 recordsLinked to original sources

bioRxiv2026

Turning Domain Expertise into Multi-Dimensional Evaluation of Biomedical AI with Karenina

Language models and agents are increasingly used in biomedicine, but current benchmarks reward correct answers even when the underlying reasoning is flawed. Here we introduce Karenina, an open-source framework that turns expert knowledge into multi-dimensional evaluations of questions, conversations and autonomous agents. Illustrated in Question-Answer pairs, multi-turn conversations and autonomous data-analysis, these dimensions together moves evaluation beyond scoring, enabling trustworthy decision-making with AI in biomedicine.

bioinformatics↗

bioRxiv2026

The Human Pleiotropic Map of GWAS Associations and Therapeutic Implications

Genetic support for drug targets substantially increases clinical success rates, establishing genome-wide association studies (GWAS) as central to therapeutic hypothesis generation. However, the same genetic evidence that reveals causal gene-disease relationships simultaneously exposes organism-level safety liabilities--a dimension requiring principled, genome-wide quantification. Here we systematically analyse 100,526 GWAS to yield 789,453 credible sets and gene prioritisations for 15,641 genes, with discovery showing no saturation as GWAS expand and increase diversity. We find that 64% of GWAS-implicated genes are pleiotropic, associated with traits across multiple diseases and showing a non-linear relationship between the degree of pleiotropy and clinical success. Highly pleiotropic genes--concentrated in immune, inflammatory, and oncogenic signalling programmes--are enriched in safety-terminated clinical programmes, mouse lethal knockouts, and cancer driver genes, establishing gene-level pleiotropy as a potential measure of genetically-informed organism-level safety liability. Protein-altering variant (PAV) support amplifies therapeutic signal (OR = 6.0), yet PAV targets show higher average pleiotropy, introducing a competing safety liability. Combining PAV support with intermediate pleiotropy (2-5 therapeutic areas) resolves this tension, yielding OR = 10.3 and relative success = 4.8--a profile already satisfied by 52 approved therapies. As GWAS continue to expand in scale and resolution, these findings lay the groundwork for increasingly sophisticated target discovery strategies that yield safer and more effective therapeutic hypotheses.

genomics↗

bioRxiv2026

Chemical Probes in Scientific Literature: Expanding and Validating Target-Disease Evidence

Chemical probes are indispensable tools for validating therapeutic hypotheses, yet their broader impact on early-stage drug discovery remains unquantified. To our knowledge, this study represents the first systematic, large-scale investigation of the chemical probe literature. By screening over 18 million articles using a high-quality dictionary of 561 chemical probes, we identified 20,000 articles mentioning a chemical probe which resulted in 5,558 unique target-disease (T-D) associations. Our analysis yields four principal findings that redefine the utility of these chemicals: First, we show that chemical probe evidence typically precedes the appearance of structured data in major knowledge bases by 1-7 years, providing a crucial lead time for target prioritisation. Second, we identified 353 T-D pairs (6.4%) with no prior evidence in the Open Targets Platform, highlighting the approachs discovery potential. Third, the application of strict novelty filters uncovered 135 new high-confidence associations between targets and diseases, revealing distinct opportunities for therapeutic repurposing in non-oncological, rare autoimmune diseases, and diseases without effective therapies due to complex biology or high treatment resistance. Finally, we demonstrate that chemical probes are essential for strengthening evidence, providing functional validation for associations previously supported only by weaker, correlative data such as RNA expression or animal models. Collectively, these findings illustrate that chemical probes catalyse early therapeutic discovery, emphasising the importance of cataloguing existing probes and identifying new ones.

bioinformatics↗

bioRxiv2024

Associations on the Fly, a new feature aiming to facilitate exploration of the Open Targets Platform evidence

MotivationThe Open Targets Platform (https://platform.opentargets.org) is a unique, comprehensive, open-source resource supporting systematic identification and prioritisation of targets for drug discovery. The Platform combines, harmonises and integrates data from >20 diverse sources to provide target-disease associations, covering evidence derived from genetic associations, somatic mutations, known drugs, differential expression, animal models, pathways and systems biology. An in-house target identification scoring framework weighs the evidence from each data source and type, contributing to an overall score for each of the 7.8M target-disease associations. However, the previous infrastructure did not allow user-led dynamic adjustments in the contribution of different evidence types for target prioritisation, a limitation frequently raised by our user community. Furthermore, the previous Platform user interface did not support navigation and exploration of the underlying target-disease evidence on the same page, occasionally making the user journey counterintuitive. ResultsHere, we describe "Associations on the Fly" (AOTF), a new Platform feature - developed as part of a wider product refactoring project - to enable formulation of more flexible and impactful therapeutic hypotheses through dynamic adjustment of the weight of contributing evidence from each source, altering the prioritisation of targets. Availability and implementationAll Open Targets code is available as open source: [https://github.com/opentargets]. This tool was implemented using React v18 and its code is accessible here: [https://github.com/opentargets/ot-ui-apps]. The tools described in the paper are accessible through the Open Targets Platform web interface [https://platform.opentargets.org/] and GraphQL API (https://platform-docs.opentargets.org/data-access/graphql-api). Data is available for download here: [https://platform.opentargets.org/downloads] and from the EMBL-EBI FTP: [https://ftp.ebi.ac.uk/pub/databases/opentargets/platform/]. Supplementary informationAdditional information on this tool can be found on the Platform documentation pages [https://platform-docs.opentargets.org/web-interface, https://platform-docs.opentargets.org/web-interface/associations-on-the-fly, https://platform-docs.opentargets.org/target-prioritisation] and training video [https://youtu.be/2A9bksboAag]. ContactAnnalisa Buniello, EMBL-EBI, buniello@ebi.ac.uk

bioinformatics↗

bioRxiv2020

Whole genome sequencing for diagnosis of neurological repeat expansion disorders

BackgroundRepeat expansion (RE) disorders affect ~1 in 3000 individuals and are clinically heterogeneous diseases caused by expansions of short tandem DNA repeats. Genetic testing is often locus-specific, resulting in under diagnosis of atypical clinical presentations, especially in paediatric patients without a prior positive family history. Whole genome sequencing (WGS) is emerging as a first-line test for rare genetic disorders, but until recently REs were thought to be undetectable by this approach. MethodsWGS pipelines for RE disorder detection were deployed by the 100,000 Genomes Project and Illumina Clinical Services Laboratory. Performance was retrospectively assessed across the 13 most common neurological RE loci using 793 samples with prior orthogonal testing (182 with expanded alleles and 611 with alleles within normal size) and prospectively interrogated in 13,331 patients with suspected genetic neurological disorders. FindingsWGS RE detection showed minimum 97{middle dot}3% sensitivity and 99{middle dot}6% specificity across all 13 disease-associated loci. Applying the pipeline to patients from the 100,000 Genomes Project identified pathogenic repeat expansions which were confirmed in 69 patients, including seven paediatric patients with no reported family history of RE disorders, with a 0.09% false positive rate. InterpretationWe show here for the first time that WGS enables the detection of causative repeat expansions with high sensitivity and specificity, and that it can be used to resolve previously undiagnosed neurological disorders. This includes children with no prior suspicion of a RE disorder. These findings are leading to diagnostic implementation of this analytical pipeline in the NHS Genomic Medicine Centres in England. FundingMedical Research Council, Department of Health and Social Care, National Health Service England, National Institute for Health Research, Illumina Inc

genomics↗

bioRxiv2018

G2P: Using machine learning to understand and predict genes causing rare neurological disorders

To facilitate precision medicine and neuroscience research, we developed a machine-learning technique that scores the likelihood that a gene, when mutated, will cause a neurological phenotype. We analysed 1126 genes relating to 25 subtypes of Mendelian neurological disease defined by Genomics England (March 2017) together with 154 gene-specific features capturing genetic variation, gene structure and tissue-specific expression and co-expression. We randomly re-sampled genes with no known disease association to develop bootstrapped decision-tree models, which were integrated to generate a decision tree-based ensemble for each disease subtype. Genes generating larger numbers of distinct transcripts and with higher probability of having missense mutations in normal individuals were significantly more likely to cause neurological diseases. Using mouse-mutant phenotypic data we tested the accuracy of gene-phenotype predictions and found that for 88% of all disease subtypes there was a significant enrichment of relevant phenotypic abnormalities when predicted genes were mutated in mice and in many cases mutations produced specific and matching phenotypes. Furthermore, using only newly identified genes included in the Genomics England November 2017 release, we assessed our gene-phenotype predictions and showed an 8.3 fold enrichment relative to chance for correct predictions. Thus, we demonstrate both the explanatory and predictive power of machine-learning-based models in neurological disease.

genetics↗