Browse Articles
4 published articles
ark:24975/genrxiv-2026-00002 · posted 2026-09-29Epidemica: An Open, Agentic-Ready Research Platform for Participatory Epidemiology Studies
Epidemica is an open-source platform for building study apps that collect high-resolution, multi-modal epidemiological data — including proximity contacts, survey responses, and, in a companion transmission engine, a simulated infection outcome — and for running interventions over that data. It is organised as a set of versioned data and protocol contracts, implemented once each in Dart client packages, an Elixir/Phoenix server, and a Python transmission model, so that a study is authored as a configuration document rather than a fork of an app. This paper describes the platform's motivation, its architecture, a reference application (a seven-day transmission game whose infections are decided by a real contact network rather than scripted), and a purposeful extension of that same contracts-first discipline to the agents that help build it: machine-readable convention files that let a coding agent work in this codebase without re-deriving rules that have already been paid for in debugging time. We report the platform's current state and discuss next steps and further directions. Epidemica is available under the Apache-2.0 license at: https://github.com/colabobio/epidemica
ark:24975/genrxiv-2026-00006 · posted 2026-09-21Rentosertib and geroprotective endpoints - an AI-assisted reanalysis and critical assessment
Two linked publications have proposed that proteomic aging clocks can serve as geroprotective endpoints in conventional drug trials. A phase 2a randomised trial of rentosertib, a TNIK inhibitor, in idiopathic pulmonary fibrosis reported a mean 12-week forced vital capacity (FVC) gain of +98.4 ml in its highest-dose arm2; a companion analysis then applied six published proteomic clocks to serum from 42 of the same participants and reported biological-age reductions of 2.7 to 3.5 years, proposing that aging and disease endpoints can be assessed simultaneously1. We recomputed every quantity we cite from the deposited supplementary tables and source data. Published headline values reproduce exactly, including the 21 of 54 significant clock comparisons, the contrast of 326 treated against 2 placebo proteins, and the four protein–FVC correlations. Six findings do not survive reanalysis. First, no between-arm contrast is reported for the trial’s lead efficacy endpoint; forming it from the published intervals gives +118.7 ml (95% CI -3.4 to 240.8, P = 0.056), and the arms differ at baseline by 0.88 standardized units of FVC in the direction of the claim. Second, all 54 clock comparisons are one-sided despite a Methods statement that tests were two-sided; the two-sided recomputation leaves 13 of 54 significant. Third, the 326-versus-2 contrast compares an interaction term with a main effect and depends on a per-term false-discovery partition; pooling the four compared terms gives 222 versus 12. The associated “mirror-image” enrichment between placebo and treated arms is weaker than a no-effect simulation of the same coding predicts (observed r = -0.622, -0.617, -0.563; simulated null median -0.690). Fourth, all six clocks were trained on UK Biobank EDTA plasma and applied to serum with plate-median normalization and, in the authors’ own words, “no bridging or anchoring controls”, so absolute biological ages are not on a calibrated scale. Fifth, a sensitivity analysis removes one placebo patient whose 12-week FVC gain was about +526 ml and one 30 mg once-daily patient, on a baseline-quality criterion stated only in a figure legend, which moves the placebo comparator down by 42 ml and grows the headline difference by 35 to 45% while leaving the highest-dose arm untouched. Sixth, none of the six clocks has a published test–retest reliability; at n = 10 the smallest detectable shift is almost exactly the test–retest standard deviation, so the reported effects are interpretable only if that unmeasured quantity is below about 2.7 to 3.5 years. We set out separately what the two studies do establish, and what a defensible geroprotective-endpoint trial would require.
ark:24975/genrxiv-2026-00004 · posted 2026-09-15FALS Therapeutics: Prompting a Blueprint for a Biotech Discovering and Developing Disease-Modifying Therapies for Amyotrophic Lateral Sclerosis Using an Agentic AI Framework
The high cost and complexity of early-stage drug discovery create profound barriers for rare-disease programs with limited funding. Sporadic amyotrophic lateral sclerosis (sALS) exemplifies this challenge: despite affecting 90-95% of the ~32,000 prevalent ALS patients in the United States, no approved disease-modifying therapy exists specifically for sALS. Here we demonstrate FALS Therapeutics, a conceptual virtual biotech blueprint built and directed entirely through iterative prompting of Claude Sonnet 4.6 (Anthropic), a commercially available agentic large language model accessed through the claude.ai conversational interface without custom software infrastructure, MCP servers, or fine-tuning. A CEO agent received a USD 5M seed mandate and instantiated four AI direct-report agents: Head of Commercial, Head of Research, Head of CMC, and Head of Clinical Development, forming a Development Committee (DC) for stage-gate governance. The virtual organization completed four sequential work-streams: commercial landscape analysis, research master plan including target selection and molecule generation decision, clinical development master planning including a patient enrichment modeling strategy and combination trial proposal, and CMC manufacturing planning across four drug modalities. The blueprint was created by the AI agent via planning documents and interactive dashboards, presentation decks, figures, tables and code. Key outputs include an ~USD 791-812M global ALS market characterization, a 12-target ranked biological long-list with TBK1 identified as a novel Tier 1 target not in the initial mandate, a fully open-source AI tool stack saving ~USD 302,000/year in licensing costs, a Phase 1-3 clinical development plan with a four-layer patient enrichment model (FALS-PEM) estimated to reduce Phase 2b sample sizes by 40-60%, a Phase 2a and 2b combo trial of two assets in the pipeline proposed by the AI agent without prompting, and a consolidated CMC budget of ~USD 35.5M across all modalities from seed to NDA. A critical design feature - the iterative gap-identification and gap-closure loop directed by the CEO agent by prompts from expert human-in-the-loop - produced 18 governance documents without additional human specification. A systematic reference verification audit - the primary methodological contribution - flagged seven categories of AI-generated claims requiring caution and identified and corrected three errors introduced by AI hallucination. This work extends earlier published work of AI agents in drug discovery by demonstrating the lower bound of what expert-guided prompting of an unmodified conversational AI, without bespoke agent infrastructure, can achieve. It can generate the full portfolio of pre-IND planning documents for a rare-disease biotech, compressing months of work into days at minimal cost but expert review and continuous feedback via prompting is still crucial for ensuring scientific and technical validity.
ark:24975/genrxiv-2026-00001 · posted 2026-09-04GenRxiv: An Open Archive for AI-Generated Research
GenRxiv is a preprint archive built specifically for research that was substantially generated or co-generated by artificial intelligence. Where existing preprint servers were designed for human-authored work and are still working out how to handle AI-generated submissions, GenRxiv starts from the opposite assumption: AI involvement is the default, not an exception to be argued for. Submissions are Markdown only, authors are identified by ORCID iD, every paper is released under CC0, and the archive is engineered to be readable by both people and machines — including autonomous agents that can prepare submission files on a human's behalf. This paper describes the motivation, design decisions, and architecture of GenRxiv, and discusses its limitations and open questions.