Research
Built on published, peer-reviewed science.
ToxIndex is the latest in a decade of shipped, peer-reviewed work. Here is the record it stands on.
- $5.8M
- in federal & EU research funding since inception
- 10+
- years building scientific AI
- 3
- shipped, peer-reviewed platforms
- 2,800+
- citations of our founder's research
Peer-reviewed
The papers behind the platform
Machine learning of toxicological big data enables read-across structure activity relationships outperforming animal test reproducibility
Luechtefeld, Marsh, Rowlands, Hartung · Toxicological Sciences, 2018 · 165(1):198–212
The RASAR paper. Models trained on a machine-readable REACH database of 9,801 substances and 816,048 study documents reached 80–95% balanced accuracy, beating the reproducibility of the animal tests they replace. It is the foundation the platform is built on.
BioBricks.ai: A versioned data registry for life sciences data assets
Insilica · Frontiers in Artificial Intelligence · arXiv 2408.17320
A package manager for scientific data. Each dataset is a versioned, reproducible ETL pipeline, so heterogeneous public data becomes analysis-ready in minutes instead of days. BioBricks supplies the curated-data foundation ToxIndex's knowledge layer draws on.
Sysrev: A FAIR platform for data curation and systematic evidence review
Insilica · Frontiers in Artificial Intelligence, 2021 · 4:685298
The systematic-review platform that proved human reviewers plus machine learning works in production — thousands of projects, teams of 100+, and over 94 million relevance predictions by early 2021. Years of real-world operation behind ToxIndex.
Public safety
Holding research transparency to account
Chemical safety is one front. The same team publishes peer-reviewed research on whether clinical-trial results reach the public at all, work we carry out under an FDA contract.
Compliance with results reporting at ClinicalTrials.gov before and after the 2017 FDAAA Final Rule
Mughal, Fu, Luechtefeld, Chiswell, Kleinstreuer, Shaw, Tidmarsh · Journal of the Academy of Public Health, 2025
A before-and-after audit of more than 24,000 federally mandated trials. The 2017 rule helped — on-time results reporting rose from 8.3% to 23.2% — yet roughly three in four trials still miss the deadline and about half never report within three years. Co-authored with scientists from NICEATM/NIEHS, Duke, and Stanford.
Sponsor-level compliance with ClinicalTrials.gov reporting requirements
Mughal, Luechtefeld, Campbell Cashman, Tidmarsh · Journal of the Academy of Public Health, 2025
The first organization-by-organization audit of trial-reporting compliance: 257 sponsors, 10,605 trials. It turns an aggregate statistic into named accountability, exposing a stark gap: industry sponsors report at 73.7% while academic medical centers manage just 25.5%.
This work is non-dilutive research, backed by NIH, NSF, FDA, and EU grants. See the full funding record.
ONTOX
Helping build the post-animal-testing toolkit
Insilica is a partner in ONTOX, a €17.2M EU Horizon 2020 consortium of 19 partners and 60+ scientists working to enable human chemical risk assessment without animal testing. We contributed three of its core AI tools — Sysrev, ToxTransformer, and ToxIndex — while BioBricks.ai supplies the consortium's big-data substrate and Sysrev powers evidence synthesis across dozens of its projects.
Read about the ONTOX projectOpen source
Work you can inspect and use
See it on your own chemistry.
Try the free version now, or book a walkthrough of the full platform on your substances.