Combining AI and Biology to Predict
Complex Toxicological Endpoints
Hybrid QSAR models integrate structural compound data with signals from multiple wet lab assays to predict complex toxicity endpoints — like chemical-induced carcinogenicity — that neither source could reliably model alone.
The Hybrid Advantage
Pure in silico tools are fast and cover thousands of compounds — but carry inherent uncertainty and have limited applicability domain. In vivo approaches are gold-standard but slow, expensive, and have ethical issues. Hybrid models combine in silico with in vitro to predict the outcome of in vivo experiments.
We implemented a hybrid model specifically designed to predict the outcome of a chemical-induced two-year rodent carcinogenicity test based on the structure of the compound and its biological response to various in vitro tests.
Representative confidence levels for toxicity endpoint predictions.
When hybrid models are the right choice
Hybrid approaches deliver the most value in specific scenarios.
Lead optimisation
Hybrid models score compounds using both structural features and biological assay signals — letting you flag and deprioritize toxic candidates early, before committing to costly in vivo studies. Fewer bad compounds make it further down the pipeline.
Mechanistic validation
Because predictions are built on biological assay readouts, the model tells you which responses are driving a toxicity signal — not just whether a compound is likely carcinogenic, but through which biological pathway.
Regulatory dossier support
Regulators accept weight-of-evidence approaches. OECD-compliant in silico predictions combined with supporting in vitro assay data create a more compelling safety narrative — and a clearer path to submission — than either source alone.
Custom model training
Have proprietary experimental data? We can train custom QSAR models on your in-house assay results — combining your wet lab IP with our in silico infrastructure.
Interested in a hybrid approach?
Contact us to discuss your compound pipeline and we'll recommend the right combination of in silico and experimental methods.