BioMap launches AI drug safety system with xTrimoTox, reports 0.9637 AUROC on liver injury
BioMap launched an AI-assisted nonclinical drug safety assessment system on November 19. It combines virtual-cell models, toxicity prediction and evaluation services, and the company reports AUROC of 0.9637 for drug-induced liver injury on a selected benchmark task.
BioMap launched an AI-assisted drug safety assessment system on November 19, combining virtual-cell models, toxicity prediction and nonclinical evaluation services. According to the company account published by Leiphone, the system is designed to help researchers prioritize candidate molecules and safety questions before committing resources to animal studies.
xTrimoTox and the "virtual monkey" approach
One core component is xTrimoTox, a toxicity-prediction model that BioMap says it built on a cell-world model. The company is also exploring virtual dosing through safety-prediction models described as "virtual monkeys," which extend predictions from cellular responses toward animal-level effects. This remains a modeling approach to be checked against experiments, not an established replacement for animal testing.
Company-reported benchmark results
BioMap reports the following figures. None have been independently reproduced.
- Drug-induced liver injury: AUROC of 0.9637 on selected Therapeutics Data Commons benchmark tasks.
- hERG-related cardiac toxicity: AUROC of 0.8918 on selected Therapeutics Data Commons benchmark tasks.
- Blinded liver-injury test: AUROC of 0.9076 across 47 drugs.
- Original Tox21 challenge test set: mean ROC-AUC of 0.8598 across 645 molecules and 12 toxicity endpoints, compared with 0.8470 for DeepTox.
These are discrimination metrics, not clinical-trial results or diagnostic accuracy percentages. They are not evidence that a medicine is safe or effective in patients. The DeepTox comparison applies to the Tox21 benchmark setting, not to safety evaluation generally.
Where it fits in the nonclinical workflow
BioMap describes support at several stages:
- Before a study: The system combines laboratory toxicity findings, small-animal data and historical evidence from larger animals or related molecules. These inform species selection, dose groups and potential target-organ risks.
- During testing: It analyzes body weight, food intake, laboratory measurements and toxicokinetic data for signals that need investigation.
- Pathology and reporting: It offers whole-slide prescreening, suspicious-region annotation, and consistency checks across source data, report tables and submission materials. These functions assist researchers and pathologists and do not replace their professional judgment.
Partners and feedback loop
BioMap says it has partnered with more than 30 contract research organizations and laboratories, and that relevant capabilities are already entering projects. In its proposed feedback loop, partner animal studies test the risks the models flag. The experimental results are then used to check and refine later predictions.
Pricing, availability terms, parameter counts and context window are not stated in the company account.
What this means
The launch is an AI-supported research workflow, not a validated replacement for animal studies or a demonstration of regulatory acceptance. The reported AUROC and ROC-AUC values show how well xTrimoTox separates toxic from non-toxic compounds on specific benchmarks. They do not show how it performs in a given drug program. The blinded 47-drug liver-injury result is the figure closest to a held-out check, and it is also company-reported. The more consequential test is the feedback loop with its 30-plus partner organizations. There, animal-study outcomes will show whether the flagged risks hold up and whether the models improve with new data.
Related Articles
Nous Research brings open-source Hermes Agent to Microsoft Teams, adds Collective Wisdom skill sharing
Nous Research says companies will soon be able to run its open-source Hermes Agent directly in Microsoft Teams, with agents sandboxed in Azure Container Apps. A new feature, Collective Wisdom, lets employees share reviewed skills with colleagues, and the shared knowledge stays inside the company.
Kimi Code CLI v2.7.0 adds native Amazon Bedrock provider with standard AWS credentials
Kimi Code CLI v2.7.0, Moonshot AI's open source coding agent, adds a native bedrock provider. It resolves the standard AWS credential chain, including named profiles and SSO, so users no longer need a separate Amazon Bedrock API key. AWS demonstrates the setup with Kimi K3.1 Pro Highspeed.
OpenAI tests ChatGPT dots with a small group of Plus users, rolls out to all Business Standard
OpenAI has started testing ChatGPT dots with a small number of ChatGPT Plus subscribers and is rolling it out to all ChatGPT Business Standard users. Dots launched in September for Pro, Business Premium and Enterprise plans. Usage allowances for the new tiers have not yet been clarified.
Databricks opens Genie ZeroOps public beta after agent observed 80,000+ customer issues in private preview
Databricks opened Genie ZeroOps to public beta on November 19, 2026, after private-preview testing. The company says the autonomous data-pipeline agent observed more than 80,000 real customer issues during that limited preview.
Comments
Loading...