About & Scientific Governance

An honest reasoning layer for drug-repurposing research.

GaiaLab turns a gene panel into a mechanism-ranked, fully-cited, testable drug-repurposing hypothesis — and holds itself accountable in public. It is independent, open, and deliberately transparent about what it can and cannot do.

Mission

Public biology already has excellent data — knowledge bases, association scores, literature, trial registries. What is scarce is trustworthy reasoning over that data: a way to go from a gene panel to a defensible, citable hypothesis without hidden assumptions or inflated claims. GaiaLab exists to be that layer — the reasoning and accountability on top of the evidence, not another database competing on source count.

We optimize for the things the research literature says actually earn a tool's adoption: reproducibility, provenance, and trust. Every claim carries its source; every prediction carries a timestamp; every benchmark is published — including the ones that do not flatter us.

What makes it different

1 · Reasoning, not just retrieval

Most platforms return scored associations and leave the reasoning to you. GaiaLab produces a narrated, mechanism-ranked hypothesis — with contradictions surfaced rather than hidden, and a proposed next experiment — so you can challenge the result, not just read a score.

2 · Accountability in public

GaiaLab timestamps every drug–disease prediction and scores it against the live ClinicalTrials.gov record, publishing the accuracy openly with naive baselines. We report our confirmed forward-prediction count honestly — including when it is zero — on a live ledger. Verification is independent of the model that generated the prediction.

3 · Depth in neglected diseases

Alongside oncology, GaiaLab goes deep on high-burden, under-served problems — such as an honestly-benchmarked Lassa fever early-warning model on real, cross-validated NCDC surveillance data — the kind of work general platforms rarely prioritize.

Scientific methodology

Each analysis runs a parallel fetch across 75+ integrated biological databases (about 72 active on a typical run, depending on available API keys), an eight-stage evidence pipeline, FDR-corrected pathway enrichment, and mechanism-aware drug scoring, cross-referenced against ClinicalTrials.gov. A structured multi-agent AI debate is available as a Pro layer. The full method is documented at /methodology, and results are reproducible — every analysis is snapshot-able with a permalink and timestamp.

Evaluation philosophy

We hold a strict line between retrospective correspondence (a matching trial exists) and prospective skill (a prediction made before the trial existed, then borne out). The first is a proxy for research relevance; only the second tests foresight, and it is the harder, honest test. Our benchmarks are shown side by side with their scope and limitations at /validation and /calibration:

Retrospective drug AUROC 0.545 (N=529) — a modest signal above the 0.50 random baseline; not a clinical predictor.
Temporal-holdout AUROC 0.90 (n=22 curated approvals) — a different, easier test; also not a clinical predictor, and not a prospective result.
Prospective forward skill — pending. Pre-registered; the first drug study is sealed (readout 2028). Confirmed forward hits so far: reported honestly, currently zero.
LassaAI AUROC 0.880 vs 0.849 naive — a ~0.03 gain that is not statistically significant at this sample size (DeLong p=0.61); seasonality-driven; the prospective season test is the genuine one.

Trial correspondence does not establish treatment efficacy. We publish what we have, not what looks best.

Limitations

Outputs are computational research hypotheses, not validated findings or clinical recommendations. Independent experimental validation is required before any therapeutic application.
Retrospective metrics overstate operational value; prospective validation is pending, not complete.
External databases and APIs degrade or rate-limit; analyses continue on partial data and say so, rather than failing silently.
The platform has not been prospectively validated, and we do not claim it has been.

Founder

Oluwafemi Idiakhoa — independent researcher, Houston, Texas, United States. Sole developer of GaiaLab and the LassaAI forecasting pipeline.

Contact: partnerships@gailabai.com · Code: github.com/oluwafemidiakhoa

Publications & research artifacts

LassaAI preprint (Zenodo, CC-BY): forecasting elevated Lassa fever transmission weeks in Nigeria — DOI 10.5281/zenodo.21122486.
Prospective drug-repurposing protocol (pre-registered, sealed): DOI 10.5281/zenodo.21420447.
Open source: the LassaAI pipeline and data are released under the MIT License at github.com/oluwafemidiakhoa/lassaai.

Independence, funding & conflicts of interest

Funding

No external, institutional, or grant funding. GaiaLab is developed independently.

Conflict of interest

The founder develops GaiaLab, a platform in which he holds a financial interest. The research pipelines described are released open-source under the MIT License.

AI assistance

Large language models are used for language editing and engineering assistance; they do not generate, analyze, or interpret scientific data or results. The founder takes full responsibility for all content.

Endorsements

No claim of endorsement by any agency, institution, or expert. Acknowledgements do not imply endorsement.

Responsible-use commitment

Research use only. GaiaLab is not a medical device and is not for diagnosis, treatment selection, or direct clinical decision-making. Do not upload protected health information (PHI), direct identifiers, or identifiable patient records. All outputs require independent expert and experimental validation before any clinical or therapeutic use.

We do not claim regulatory approval or compliance certifications. Where a limitation exists, we state it. Where a result is uncertain, we say so. That honesty is not a caveat bolted on — it is the point of the platform.