Verifiable Agentic Science on Public Infrastructure

Baker, D.; Cain, S.; van den Beek, M.; Chilton, J.; Callan, D.; Weaver, S.; Coraor, N.; Beavers, K.; Goecks, J.; Schatz, M. C.; Gruening, B.; Kosakovsky Pond, S. L.; Nekrutenko, A.

doi:10.64898/2026.10.26.761864

bioRxiv · 10.64898/2026.10.26.761864

Verifiable Agentic Science on Public Infrastructure

Baker, D.·Cain, S.·van den Beek, M.·Chilton, J.·Callan, D.·Weaver, S.·Coraor, N.·Beavers, K.·Goecks, J.·Schatz, M. C.·Gruening, B.·Kosakovsky Pond, S. L.·Nekrutenko, A.

View original source ↗Export metadata ↓

Abstract

Autonomous artificial intelligence agents are transforming the practice of science. Across every discipline, researchers are handing data analysis to coding agents that can design and execute complex studies from a single prompt. Yet this revolution carries a threat. When agents improvise in local environments, they leave no verifiable record, ultimately risking a wave of irreproducible science. Solving this crisis requires moving agentic science onto public infrastructure where every action leaves an examinable trail. Here we connect autonomous coding agents to shared computing resources, turning ephemeral prompts into auditable scientific records. After showing that 30 unguided local agents splinter into 18 incompatible pipelines on the same dataset, we conduct an agentic reproduction and discovery study on a large scale---reproducing a large, diverse array of studies with every workflow, history, and intermediate dataset publicly accessible. This complete provenance trail exposes unstated modeling choices, reference mismatches, and subsampled tutorial inputs in published studies, and establishes a new foundation where human scientists can understand, steer, and trust AI-driven discovery.

Explore related subjects

bioinformatics Preprint

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Baker, D., Cain, S., van den Beek, M., Chilton, J., Callan, D., Weaver, S., Coraor, N., Beavers, K., Goecks, J., Schatz, M. C., Gruening, B., Kosakovsky Pond, S. L., Nekrutenko, A.. 2026-10-28. Verifiable Agentic Science on Public Infrastructure. https://doi.org/10.64898/2026.10.26.761864

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

Carrier-Limited Release Risk in Electrospun Nanofibers for Regenerative Endodontics: Constraint Discovery from 25 Formulations

Regenerative endodontics needs an intracanal carrier that keeps disinfecting across a dressing interval that clinicians schedule at one to four weeks, without harming the stem cells that rebuild the pulp. We ask what one completed 25-formulation electrospinning campaign of poly(vinylpyrrolidone) (PVP) fibers carrying metronidazole, ciprofloxacin, minocycline and Ca(OH)2 can decide, and use machine learning for constraint discovery and decision support rather than as a new predictor. After an audit against the source thesis, logistic regression predicts fine-fiber formation at 96% leave-one-out accuracy, but a one-threshold rule does as well, the gain over the 80% majority baseline is not significant (exact McNemar p = 0.125), and accuracy falls to 84% when mixture loading is encoded per component. Diameter models interpolate within formulation families (ridge MAE 193 nm, 95% CI 96 to 317 nm) and fail across them (R^2 < 0). Because the campaign never measured release, we assemble a same-drug, cross-carrier evidence table: 17 bounded metronidazole and ciprofloxacin observations from seven studies, with two-sided censoring. Water-soluble carriers (PVP, PVA) completed release within 80 h (PVP within 8 h); no water-insoluble monolithic carrier reached 90% within its observation window, releasing 1.5 to 50%; both within-study comparisons separate at matched loading; and blend, core-shell and thick-mat architectures released over 3 to 28 days. The PVP mats therefore carry a carrier-limited release risk, not a proven ceiling, and one dissolution experiment can confirm or overturn it. A censoring-aware screen keeps three necessarily and one possibly nondominated formulation, and fixed-budget replays show that uncertainty sampling finds failed formulations sooner, except the one it is confidently wrong about.

bioinformatics↗

bioRxiv2026

CellMethMarker provides an integrated database of human cell- and tissue-type-specific DNA methylation markers

Cell- and tissue-type-specific DNA methylation markers provide epigenetic signatures of cellular identity and are widely used to infer cellular origin from bulk and single-cell methylomes. Existing resources mainly emphasize disease-associated biomarkers, methylome atlases, genomic annotations or RNA-based cell markers, leaving an unmet need for an integrated region-level resource dedicated to such markers. Here we present CellMethMarker (http://gu-zhanglab.hfcas.ac.cn/CellMethMarker), a curated database integrating 347,514 de novo markers derived from uniformly processed whole-genome and reduced-representation bisulfite sequencing datasets, 2,203,507 literature-derived marker records and harmonized metadata for 2,810 samples. CellMethMarker enables cell- and tissue-centered marker retrieval, coordinate-based search, genome-browser visualization, interactive methylation analysis and data download. In an independent single-cell methylome atlas, these markers recover the expected cell-type-associated methylation patterns, and a platform-matched reference built from them lowers the mean deconvolution error from 0.365 to 0.144. CellMethMarker therefore links region-level DNA methylation variation to cellular identity and supports methylome-based cell-type annotation and deconvolution.

bioinformatics↗

bioRxiv2026

IG-28 enables identification and quantification of high-confidence isoforms from large bulk and single-cell full-length cDNA datasets

The technology to sequence full-length transcriptomes from bulk and single-cell samples has rapidly advanced. As a result, full-length cDNA datasets now reach billions of reads in size which is beyond the capability of most established isoform identification tools. Here, we introduce IG-28 which addresses this challenge with its capability to process and analyze large fl-cDNA datasets fast and using minimal computational resources. In contrast to its predecessor Mandalorion, IG-28 is not limited to bulk fl-cDNA but is capable of parsing, processing, and analyzing single-cell fl-cDNA datasets. It does so in a lightweight, flexible manner, integrating seamlessly with established scRNA-seq analysis workflows. We benchmarked IG-28 against a variety of established and new isoform identification tools using several bulk and a complex single-cell fl-cDNA dataset and found that IG-28 consistently outperformed the other tools by identifying high-confidence public isoforms.

bioinformatics↗