The Pharmacogenomics Journal

Motivation: Genetically supported drug targets are an estimated 2.6 times more likely to succeed clinically, yet a disease association alone does not show which protein to target, in which direction, or what other effects may follow. This currently requires chaining separate tools. Results: drugMR is a Nextflow pipeline that, from GWAS and pQTL summary statistics, runs cis-Mendelian randomisation, colocalisation, SMR-HEIDI with bulk and/or single-cell QTLs, three-trait colocalisation and a phenome-wide screen for adverse effects and repurposing. Datasets are added through a manifest, every step runs in containers on a laptop, cluster or cloud, and results are explored in a dashboard. Applied to schizophrenia with brain pQTLs, drugMR tested 2,019 proteins in 4 h 19 min and prioritised 58 proteogenomic targets, 22 with multi-omic support; 4 had potential adverse effects. Availability and implementation: drugMR: https://github.com/guillermocomesanacimadevila/drugMR; documentation: https://guillermocomesanacimadevila.github.io/drugMR/. Demo data: https://doi.org/10.5281/zenodo.22917384.

Computational safety evaluation of phytochemical libraries commonly relies on models trained on experimentally tested compounds and applied to much larger and chemically diverse collections. An important but often overlooked question is whether predicted compounds are adequately represented within the experimental chemical space underlying these models. Here, we examined this question for the Indian Medicinal Plants, Phytochemistry and Therapeutics (IMPPAT) database using experimentally characterized datasets for three anti-target endpoints: hERG blockade, cytochrome P450 (CYP) inhibition, and pregnane X receptor (PXR) induction. After structure standardization, 14,956 unique IMPPAT compounds were compared with 8,877 hERG, 4,905 CYP, and 4,138 PXR reference compounds using consensus nearest-neighbor similarity based on Morgan and atom-pair fingerprints. At a Tanimoto similarity threshold of 0.80, only 0.92%, 0.30%, and 0.46% of IMPPAT compounds had close measured analogs in the hERG, CYP, and PXR datasets, respectively. Relative to the internal nearest-neighbor distributions of the reference datasets, a typical IMPPAT compound lay near the 1st percentile, indicating substantial chemical-space separation. Compounds with close experimental analogs also showed distinct molecular characteristics, with descriptor-based logistic regression yielding cross-validated AUC values of 0.77-0.80. Plant-level enrichment was markedly reduced after excluding widely distributed metabolites and was further restricted when the nearest measured analog was required to have pIC50 or pEC50 [≥] 5. The remaining signals were concentrated in a limited number of taxa, including Rauvolfia, Sinopodophyllum/Dysosma, Cinchona, and Catharanthus roseus. Overall, most IMPPAT phytochemicals lack close structural representation among currently measured anti-target compounds. Computational safety predictions for these molecules should therefore be interpreted as extrapolations and accompanied by explicit measures of structural support. Absence of a close experimental analog should be regarded as a gap in available evidence rather than evidence of safety.

PurposeTo describe the development of - and outcomes from - an institution-level initiative supporting access to innovative genetic medicines information and opportunities. MethodsWe developed and implemented a Genetic Medicines Patient Assessment Pathway for rare genetic conditions at a large public hospital, with multidisciplinary teams collaborating to evaluate a patients molecular and clinical amenability to experimental and/or to-be-developed genetic medicines (primarily nucleic acid therapeutics). A Genetic Medicines Consult Clinic operating in parallel offered no-cost specialized consultations and counseling to families regarding genetic medicines inquiries and Pathway outcomes. We retrospectively reviewed submissions, decisions, and outcomes over a >2-year period. ResultsA total of 63 submissions were reviewed in the Patient Assessment Pathway. Most submissions were classified as eligible or potentially eligible for at least one genetic medicines approach (42/63; 67%). Recommendations generated 87 discrete actions, such as outreach to potential collaborators (n = 28), offering participation in existing research studies (n = 23), or referral to the Consult Clinic (n = 19). Two patients have been approved for therapeutic program initiation or clinical trial enrolment. ConclusionFindings to date support feasibility and high uptake of a structured, multidisciplinary, medical genetics-led model for assessing patient-level opportunities for genetic medicines.

The majority of antidepressant and antipsychotic medications are underscored by their common weight-related adverse drug events; medication-induced weight gain is a central hindrance to psychiatric treatments because it impacts patient compliance and results in a cascade of secondary comorbidities with negative health outcomes. Here, we use network pharmacology approaches to characterize the multiple effects of drugs and explore the mechanisms by which drug-molecule interactions lead to adverse drug events. We first construct a robust interactome containing 17,293 proteins (network vertices) and 408,883 weighted interactions (network edges) and identify communities of interconnected vertices using a modularity optimization clustering algorithm. We link 33 antidepressants and 26 antipsychotics into this network via their drug-protein interactions and implement network measures of cosine similarity, shortest path distance, and connectivity significance to determine which communities are most proximal to the drugs. We identify a subnetwork of 33 (out of 536) communities enriched for genes associated with adiposity traits to test for connectivity significance to the drug vertices. Through an additional random walk analysis, we triangulate consistent links between mental health drugs and molecular pathways, including GPCR, PI3K-Akt, leptin, and insulin signaling, as well as adipogenesis. We highlight specific components within these curated pathways and map the crosstalk between them. While we demonstrate our approach in the context of mental health drugs and their weight-related side effects, our study design is broadly applicable to a wide range of drug classes, drug events, and traits of interest.

Background and Objective: Prescribing multiple drugs simultaneously, known as polypharmacy, is increasingly common in managing chronic diseases such as diabetes and cardiovascular conditions. Studies show that roughly 40% of elderly patients are on five or more medications, putting them at significant risk of adverse drug-drug interactions (DDIs) [1]. While computational methods based on Graph Neural Networks (GNNs) have shown strong potential for predicting these interactions from a drugs molecular structure, most existing models are limited in two important ways: they rely on attention mechanisms that cannot adapt to the unique context of each atom in a molecule, and they only predict whether an interaction exists rather than how dangerous it might be. Methods: We propose HMVA-DDI, a two-stage deep learning framework that addresses both shortcomings. In the first stage, our model predicts whether two drugs will interact. If it detects an interaction, the second stage classifies it into one of three clinical severity levels: minor, moderate, or major. To better capture molecular structure, we replace the standard attention mechanism with a more powerful version called Graph Attention Network v2 (GATv2), which considers each atoms specific context when learning its representation. We also introduce a technique to prevent information loss in deeper network layers, and we tailor the input features separately for each prediction task. Results: Evaluated on two standard benchmarks, DrugBank 5.0.3 [2] and DDInter [3], HMVA-DDI achieves an AUROC of 0.9819 and AUPRC of 0.9759 on binary interaction detection, a marked improvement over the baseline. For severity classification, our best model variant (qcSvr4) reaches an overall AUROC of 0.9444, with a particularly large improvement in detecting low-severity (Minor) interactions, a historically difficult class. Conclusion: HMVA-DDI bridges the gap between computational drug research and clinical usefulness by not only predicting interactions accurately but also indicating interaction severity. This two-level system can help healthcare professionals prioritize warnings, reducing the alert fatigue problem where doctors are overwhelmed by too many non-critical notifications.

Background: Genetic and observational studies have identified metabolic factors potentially contributing to the risk of Major Depression (MD). However, studies to date have concentrated on common genetic variants, many of which are in linkage disequilibrium. This makes it difficult to localise variants to a single gene. To overcome this, we used short read whole-genome sequenced datasets (srWGS) to conduct a targeted rare variant association analysis of the FADS region, which has previously been implicated in metabolic effects on MD. Methods: Using srWGS data from 740,373 individuals from All of Us (AoU) and the UK Biobank (UKB) we tested for associations between rare variants in the FADS region and MD. Additionally we conducted association tests between rare FADS variants and five blood-based metabolites implicated by previous studies of MD. Finally, we applied Mendelian randomisation to identify causal effects of metabolites on MD using rare-variants. Results: We found FADS rare variants were significantly associated with all tested metabolites, but not with MD. Using Wald-Ratio MR we found a putative causal effect of Docosahexaenoic acid (DHA) on MD using a rare-variant instrument in the FADS2 gene. Conclusion: Our study represents the largest rare-variant analysis of MD to date. However, the lack of associations between FADS rare-variants and MD suggests our analysis is underpowered, or that genetic effects on MD are conferred substantially by higher frequency variants. Rare-variant MR suggested a causal effect of DHA on MD, and implicates the FADS2 gene.

Background: Preeclampsia (PE) is a hypertensive pregnancy disorder with contributions from both maternal and fetal genetic factors, but how both genomes regulate placental transcription remains incompletely understood. Results: We analyzed placental 5'-end RNA sequencing data from 59 pregnancies in the FINNPEC cohort and performed separate maternal and fetal cis-eQTL analyses using gene-level expression and promoter-level expression defined as Transcript Far 5' Ends (TFE). Genotype-by-PE interaction models were used to identify genetic effects that differ according to PE status. No gene-level interaction eQTLs remained significant after multiple-testing correction, although suggestive signals included LILRB2 in the maternal and LPCAT3 in the fetal analyses. In contrast, we identified two significant syntenic promoter-level interaction eQTLs: rs12504642-TFE61820 at the FHDC1 locus in the maternal analysis and rs1557150-TFE9032 at the PPIAP31 locus in the fetal analysis. Conclusion: These findings suggest that PE-dependent genetic interaction effects may be detectable at specific transcript promoter regions even when they are not captured at annotated gene levels.

Polygenic risk scores (PRS) are rapidly developing, enabling estimates of genetic risk for developing psychiatric disorders. Psychiatric PRS are also increasingly entering consumer and clinical settings. Yet, little is known about how adults with psychiatric disorders perceive their potential use. Here, we conducted semi-structured interviews with 29 adults with psychiatric disorders to examine their perspectives on hypothetically receiving psychiatric PRS for themselves and their current or future children. Key themes emerged around perceived utility, motivations and deterrents to genetic testing, interpretation of PRS, and anticipated responses to results. In particular, participants expressed hope that PRS could support self-understanding, inform better psychiatric care, and enable earlier recognition of difficulties for their children. Participants however raised multiple concerns, including those of privacy, stigma, discrimination, eugenics, and the risk of self-fulling prophecies. Together, these findings offer critical empirical evidence to inform guidance on the responsible use of psychiatric PRS in clinical and consumer settings.

Background: Huntington's disease (HD) is an inherited neurodegenerative disorder primarily associated with motor and psychiatric manifestations. No disease-modifying treatment currently exists, with vesicular monoamine transporter type-2 inhibitors (VMAT2i's) being the only FDA-approved therapy for motor features in HD. Previous studies evaluated the antichorea effects of VMAT2i's but did not address its effects on non-choreiform features. Objectives: This study aimed to characterize the effect of VMAT2i's on hyperkinetic, hypokinetic and total hypokinetic subscores of the UHDRS motor assessment in patients with motor-manifest HD from the Enroll-HD research platform. Methods: Data from 622 participants were analyzed. Linear mixed effects models were constructed to compare the rates of change of choreiform and non-choreiform motor scores before and after the initiation of a VMAT2i. We calculated and plotted the direction, magnitude and significance of VMAT2i administration on the rates of change of motor subscores. A similar analysis was performed investigating the effect of antipsychotic medications on motor subscales in HD. Results: VMAT2i's reversed progression of hyperkinesis but significantly increased the rates of change in the hypokinetic and total hypokinetic subscores. Antipsychotics also displayed efficacy in reversing progression of chorea but did not significantly influence progression of hypokinesis. Conclusions: The initiation of VMAT2i's was associated with an improvement in hyperkinesis but was associated with worsening of other motor features of HD. Hypokinesis increases progressively in HD and is correlated with functional impairment and global decline. Therefore, the use of VMAT2i's to treat motor features in HD should be a nuanced decision, considering its multipronged impact.

Background: Pharmacogenomic point-of-care testing (PGx POCT) enables genotype-informed prescribing at the bedside, but implementation is limited by the lack of suitable reference materials and standardised external quality assessment (EQA). Methods: Fixed lymphoblastoid cell lines were formulated as swab-based materials at defined cellular inputs for MTRNR1 POCT. Materials were evaluated using Genedrive MTRNR1 assays, quantitative PCR, acceptance quality limit batch release, workflow perturbation testing and stability studies, then deployed in an international EQA programme. Results: Materials were homogeneous, workflow-compatible and reproducible, with 100% correct genotype calls at the lowest intended manufacturing cell load. Seven EQA surveys yielded 253 participant results, with full-mark rates between 87.5 and 100%, and six test failures. Most failures occurred with low cell-input materials near the assay limit of detection. The materials supported genotype concordance, performance monitoring and identification of workflow and sample-related variability. Conclusions: Workflow-compatible cell-based reference materials combined with EQA provide a scalable framework for quality-assured PGx POCT beyond specialist laboratories. International MT-RNR1 PGx POCT EQA programme deployment demonstrated feasibility in decentralised settings. The framework may provide a model for the implementation, verification and ongoing quality assessment of PGx POCT. Further evaluation across platforms, analytes, settings and participant groups is required to establish broader applicability.

Abstract Polycystic ovarian syndrome and Type 2 diabetes mellitus are complex multivariate diseases that share well-established clinical pathophysiology, yet the molecular basis of their overlap is unknown. Prior bioinformatics studies have identified common differentially expressed genes between these two conditions. However, they have not accounted for the directionality of the gene expression changes observed, potentially obscuring any biological distinctions they may have. This study applied a direction-aware approach to classify molecular signatures between PCOS and T2DM as concordant or discordant. Transcriptomic analysis of GEO datasets GSE138518 (ovarian granulosa tissue, PCOS) and GSE25724 (pancreatic islet tissue, T2DM) identified 225 and 1,302 DEGs, respectively. Venn diagram analysis showed that only three genes (SLC6A8, RGS4, and SORL1) were shared between PCOS and T2DM from the total of 1,527 genes, and all 3 genes were regulated in opposing directions. Disease gene retrieval from the Comparative Toxicogenomics Database, Online Mendelian Inheritance in Man database, and GeneCards showed 214 shared disease-associated genes, with 311 genes unique to PCOS and 496 genes unique to T2DM. Protein-protein interaction construction using STRING (version 12.0) identified 70 interacting nodes. CytoHubba analysis across 6 scoring methods identified 18 high-confidence hub genes, including INS, BCL2, MTOR, LEP, MFN2, and PIK3CD. Functional enrichment analysis identified biological processes including immune activation, apoptosis, and insulin signaling, which were confirmed as the main pathways in KEGG pathway analysis. These findings show that while PCOS and T2DM share limited DEGs, they share common pathogenic pathways.

Open-loop insulin pump therapy depends on clinical expertise for dose initiation and titration. We developed a long short-term memory-based system, using 1738 treatment courses from 1641 patients with type 2 diabetes for initial prescribing and daily basal and bolus titration. Patient-grouped five-fold validation showed closer agreement with recorded prescriptions for subsequent titration than for initiation. Prospective evaluation in 97 patients from the model-directed arm of a randomized trial yielded a median derived time in range (3.9-10.0mmol/L) of 84.5%, with low hypoglycemic exposure. Favorable control extended to previously treated patients. Higher mean glucose during the first 24 hours was associated with poorer glycemic control thereafter. The association between distance from development data and glycemic outcomes was stronger when early glucose responses were included alongside baseline characteristics. These findings support model-directed pump dosing and highlight early treatment response as a clinically relevant signal for assessing applicability and identifying patients who may require additional oversight. Trial registration: ClinicalTrials.gov, NCT07066891.

Background People diagnosed with severe mental illnesses (SMI) including bipolar disorder, schizophrenia and psychosis are disproportionately exposed to adverse social determinants of health, contributing to premature morbidity and mortality. Social prescribing aims to address social risk factors for poor health by linking patients in primary care to community activities. However, whether people with SMI access social prescribing through primary care remains unexplored. Aims To describe prevalence of, and sociodemographic characteristics associated with, social prescribing consultations in patients with SMI in primary care. Methods Using the UK Clinical Practice Research Datalink, we identified patients aged 18+ with a diagnosis of SMI between 2004-2019. We calculated the percentage of patients with SMI who received a social prescribing consultation between 2019-2023 and used logistic regression to examine associations between sociodemographic, health-related factors and social prescribing consultations and refusals. We used fixed effects analysis to determine whether a new diagnosis of SMI made between 2019-2023 was associated with an increased odds of receiving a social prescribing consultation. Results Among 112,334 patients with SMI, 0.8% received a social prescribing consultation in 2019 increasing to 9.5% in 2023. Between 2020-2023, older age (OR=1.18-2.34), being female (OR=1.16-1.31) and living in more deprived areas (OR=1.14-1.49) were associated with higher odds of receiving a social prescribing consultation amongst patients with SMI. Across all years, patients with SMI and depression (OR=1.32-1.92), diabetes (OR=1.19-1.35) or obesity (OR=1.17-1.54) also had increased odds. In 140,514 patients with time-varying SMI status, a new diagnosis of SMI was associated with higher odds of receiving a social prescribing consultation (OR=1.39, 95% CI 1.26-1.53). Conclusion Since 2019, social prescribing consultations among patients with SMI in primary care have increased and are reaching those in areas of higher deprivation and with greater health needs. Strategies to remove barriers to consultations among younger adults and males are still needed

In 1998 we predicted percent human intestinal absorption (HIA) for 86 drug and drug-like compounds from six molecular descriptors chosen by a genetic algorithm with a small neural network serving as the fitness function. That result rested on one division of the compounds into three sets: training, cross-validation and external prediction. Repeating the division shows it was a rather fortunate one. An extraction is one random draw from a wide distribution, so we repeat every experiment 250 times and report the result as a distribution. Across 250 stratified extractions, 14 match the published cross-validation error and only one matches the published external prediction error. Reconstructed this way, the published six descriptors give a mean extraction error roughly twice the external prediction value we reported in 1998 (Table 1). We then applied the same six descriptors to a larger set of 780 compounds from a public compilation. Under the 1998 network they give 22.50 %HIA units on training compounds against 25.77 on extracted compounds, whereas on the 86 they give 13.43 against 30.39. More compounds close the gap between fitting and prediction. Four datasets are modeled here: 780 small molecules with HIA measured in man, 3,771 small molecules and 1,281 cyclic peptides measured across Caco-2 cell monolayers, and 7,298 cyclic peptides measured across artificial membranes. Each is its own prediction problem, and one procedure is applied to all four. A genetic algorithm chooses which descriptors to use and how many, and a random forest grades each choice on compounds its own trees never saw. The descriptors chosen most often across the 250 extractions form one consensus set for each dataset. For absorption, 25 consensus descriptors under a random forest reach an extraction rms of 20.47 %HIA units at R 0.76. On the same compounds and the same extractions, the six 1998 descriptors under the 1998 network reach 25.77 at R 0.59. The four models are served at predicthia.ai, where a structure can be drawn or pasted, scored against its training set for similarity, and the training data with fitted models downloaded. Each model served is its fixed consensus set fitted on every compound in its dataset; the accuracy quoted beside a prediction comes from evaluating that same set across 250 extractions (Table 3). The fixed sets hold 25 descriptors for absorption, 55 for Caco-2 small-molecule permeability, 50 for Caco-2 peptide permeability and 25 for artificial-membrane permeability. Keywords: human intestinal absorption; QSPR; genetic algorithm; descriptor selection; cyclic peptides; membrane permeability; oral bioavailability; repeated extraction

A compound's toxicity can be gauged from the toxicities of the compounds it resembles. We showed that in 2003 for rat oral LD50: the compounds in a reference set of 13,645 that most resemble the query form a consortium and vote on its toxicity. This work returns to that method twenty-three years later, on public data, because the reference set used in 2003 is no longer public. Rebuilt on 11,989 curated rat oral LD50 values, the method answers for 67.0% of the held-out compounds carrying an exact value, and 92.1% of those answers land within one log unit; in 2003 the figures were 94% within one log unit on 57.3% of compounds. We then ran it over 28 further endpoints, each recording whether a compound showed a particular toxicity, from cytochrome inhibition to mutagenicity to clinical toxicity: 149,224 measurements on 34,162 compounds. The method answers only when a compound has close enough neighbors, and on those it correctly identifies a median 72.9% of the compounds that showed the toxicity and 96.5% of those that did not. Separately we asked whether a reverse screen carries information about these endpoints beyond molecular size and lipophilicity. We search a compound against a published index of 27,910 ligands solved inside proteins and count the protein families those sites belong to. A compound whose matches reach no protein family inhibits CYP3A4 2.5% of the time; one whose matches span sixteen or more families inhibits 51.2% of the time. Added to heavy atom count, molecular weight and cLogP, the screen raises the mean area under the ROC curve from 0.72 to 0.76 across the 28 endpoints. The OpenADMET consortium has since measured CYP3A4 inhibition on 2,335 compounds, and the same pattern holds. Compounds that resemble known binders of sixteen or more protein families inhibit CYP3A4 24.6% of the time; compounds whose matches reach no protein family inhibit it 7.3% of the time. The more protein families a compound resembles binders of, the more likely it is to inhibit CYP3A4. This association is strong enough to act on before anything is made. Scoring a library on how far its compounds reach across protein families, together with their size and lipophilicity, and setting aside the riskiest fifth removes two thirds of the androgen receptor binders and about half of the CYP3A4 inhibitors. Scoring needs only a two-dimensional structure, with no docking and no conformer generation at query time. Interactive scoring, the data and the pool are at toxpred.ai. Keywords: toxicity prediction; structural similarity; nearest neighbors; reverse screening; target breadth; promiscuity; LD50; cytochrome P450; virtual library triage

Background: Clinical decision support (CDS) systems may improve chronic pain management by providing guideline-based recommendations within clinical workflows. However, low clinician use can limit their effectiveness. Understanding the association between CDS use and guideline-recommended care may help distinguish implementation challenges from limitations in CDS functionality. Objective: To estimate the association of OneSheet, a CDS system for chronic pain management, and guideline-recommended care when the system was used during patient encounters. Methods: We conducted a per-protocol secondary data analysis of data from a pragmatic randomized clinical trial conducted from October 2020 to May 2022 across 25 primary care clinics at 2 academic health systems in Indiana and North Carolina. The parent trial enrolled 137 primary care clinicians. This analysis included 145511 eligible chronic pain encounters and 15480 long term opioid therapy (LTOT) encounters that occurred after OneSheet implementation. OneSheet use at a patient encounter was the exposure of interest. Propensity score matching was used to create comparable groups of encounters with and without OneSheet use. Outcomes included documentation of pain-related goals, documentation of pain and function using the Pain, Enjoyment of Life and General Activity (PEG) scale, prescription drug monitoring (PDMP) review, urine drug screening (UDS) orders, and naloxone orders. Results: OneSheet was used in 959 of 145511 chronic pain encounters (0.7%) and 582 of 15480 LTOT encounters (4.0%). In matched analyses, OneSheet use was associated with increased documentation of pain-related goals (relative risk (RR)=11.70 (95% CI 4.25-32.50) and PEG score documentation (RR=9.70 (95% CI 4.14-22.31). Among LTOT encounters, OneSheet use was associated with increased UDS ordering (RR=1.52 (95% CI 1.40-1.65) and PDMP review (RR=1.06 (95% CI 1.03-1.09), but decreased naloxone ordering (RR=0.13 (95% CI, 0.02-0.90). Conclusions: OneSheet use was associated with several guideline-recommended chronic pain management actions, suggesting that they system may support care quality when clinicians use it. However, use was rare, indicating that increasing routine use through better workflow integration may be necessary to achieve population-level improvements in chronic pain care.

Myotonic dystrophy type 1 (DM1) is caused by a CTG expansion in the 3' untranslated region of the DMPK gene that results in the expression of expanded CUG RNA which trap splicing factors resulting in widespread alternative splicing dysregulation. A panel of these splicing events, or splicing index (SI), is currently used to measure splicing dysregulation in DM1 patients. Despite significant interest in splicing, little is known about its progression over time in DM1. The objective of this study was to evaluate the changes in SI over 3 years in a well-phenotyped cohort of DM1 participants and identify factors contributing to these changes. RNA was extracted from vastus lateralis (VL) biopsies of 20 DM1 participants taken 3 years apart and the SI score was measured using targeted RNA sequencing. Baseline SI scores correlated with percent predicted muscle strength of the ankle dorsiflexors, knee extensors and hand grip. On average, SI scores worsened over 3 years, but the progression varied markedly among participants. Adult/juvenile phenotypes and lower baseline SI scores were among the factors associated with higher progression of SI scores over time. These findings emphasize the importance of integrating clinical phenotype and baseline molecular severity when interpreting longitudinal SI change, particularly in the context of clinical trials.

Large language model (LLM) agents are increasingly capable of carrying out autonomous computational research, but it remains unclear whether they can develop molecular modeling methods that compete with strong human-developed approaches. Here, we evaluate autonomous method development across four settings: Therapeutics Data Commons (TDC) ADMET tasks, the OpenADMET ExpansionRx Challenge, the activity prediction track of the OpenADMET PXR Induction Challenge, and the Practical Molecular Optimization (PMO) benchmark. Agents followed the prescribed data splits, metrics, and evaluation protocols for each benchmark. On TDC, the best agent-developed score across configurations improved on the leaderboard reference on 18 of 21 tasks and on the combined leaderboard and peer-reviewed reference on 17, while individual Codex configurations improved on the combined reference on 13 tasks. On ExpansionRx, five-agent and single-agent Codex configurations achieved overall scores corresponding to fourth and sixth place relative to the published final leaderboard, with the five-agent configuration reaching top-five performance on seven of nine endpoints. On PXR, five-agent Codex achieved a mean absolute error slightly outperforming the best published challenge entry. On PMO, five-agent Codex developed a molecular optimizer that outperformed existing methods evaluated under the benchmark's standard setup; other methods obtained higher reported scores under modified conditions, including additional molecular pretraining data or a different oracle-call budget. Together, these results show that current LLM agents can autonomously develop high-performing molecular prediction and optimization methods across substantially different drug-discovery settings, in several cases matching or exceeding leading human-developed approaches.

Background/Aims: Pre-stroke depression has been shown to be associated with worse functional outcomes and depression after stroke. However, it is not currently well understood whether pre-stroke mood disorders (e.g., depression, anxiety, mood and psychotic disorders, substance use disorder) are associated with withdrawal of life-sustaining treatment (WOLST) after a severe stroke. Methods: This was a retrospective cohort study within the Tufts Vascular Neurology Registry of patients hospitalized with severe acute ischemic stroke (AIS, defined as NIHSS [≥]10) or severe nontraumatic intracerebral hemorrhage (ICH, defined as ICH score [≥]2) between 2018 and 2022. Multivariate logistic regression was performed to identify predictors of WOLST. Results: This study included 405 patients with severe stroke. Patients who opted for WOLST were more likely to be older, female, have AIS, and experience more severe strokes. In an adjusted analysis (basic demographics, stroke type), a pre-stroke mood disorder was associated with WOLST (OR 2.15, 95% CI 1.11-4.15). In models stratified by stroke type (AIS vs. ICH) and adjusted for stroke severity, WOLST was associated with age in both groups and with mood disorder in patients with ICH. The WOLST event rate was higher in the ICH group (19% vs 11%). Conclusion: Pre-stroke mood disorder was associated with increased rates of withdrawal of care after severe stroke in this patient population. Opportunities to actively treat depression may be useful to include in goals of care discussions to mitigate the potential impact of this factor in decision-making. Consistent with prior studies, we observed a higher proportion of WOLST in older, female, and ICH patients. Future studies regarding the quality and content of goals of care discussions may help elucidate the differing effects of pre-stroke mood disorders on WOLST by stroke type seen in this study.

Observational studies suggest an association between COVID-19 and acute myocardial infarction (AMI), however, whether this relationship is causal remains unclear. We conducted a multi-ancestry Mendelian randomization (MR) study across European, East Asian, South Asian, and African populations, evaluating three COVID-19 phenotypes: very severe COVID-19, hospitalized COVID-19, and general infection association with AMI. GWAS summary statistics were obtained from the COVID-19 Host Genetics Initiative Round 7 and UK Biobank/Biobank Japan for AMI. Inverse-variance weighted (IVW) MR was the primary method, with fixed- or random-effects models selected according to heterogeneity test. MR-Egger, weighted median and weighted mode were conducted as sensitivity analyses. In European ancestry, genetic liability to very severe COVID-19 and hospitalized COVID-19 was slightly inversely associated with AMI risk, but without statistical significances across sensitivity analyses. Hospitalized COVID-19 showed a suggestive but non-significant inverse association in South Asian ancestry. No significant associations were identified in East Asian or African ancestries. Our findings did not support a robust causal association between COVID-19 susceptibility or severity and AMI risk across ancestries, suggesting that the observed cardiovascular burden may be driven more by acute pathophysiological mechanisms than shared genetic architecture.