Two ways to extract value
from in silico predictions
In silico predictions carry inherent uncertainty. The key is knowing how to use them effectively. Toxometris is designed for both use cases.
Mechanistic interpretation of predictions
When working with a limited number of compounds, the most value comes from understanding why a prediction was made — not just what the prediction is.
For each compound, our reports include read-across analysis comparing your molecule to the three most structurally similar compounds with known experimental values, plus structural alert analysis for genotoxicity endpoints.
Lead optimisation
Understand which structural features drive toxicity flags and iterate intelligently.
Regulatory submission support
Get OECD-compliant predictions with mechanistic narrative ready for inclusion in safety dossiers.
Expert report add-on
Combine automated predictions with written interpretation from medicinal chemists and toxicologists.
Lower score = higher likelihood of pharmaceutical acceptance
Risk Score bulk screening
When working with hundreds or thousands of compounds, reviewing every endpoint for every molecule is impossible. The Risk Score collapses your entire ADMET profile into a single number.
Submit your full compound library. The Risk Score ranks every compound by its likelihood of acceptance as a pharmaceutical — so your team can focus resources on the most promising candidates.
Consensus models across multiple molecular representations
Advanced algorithms with mechanistic interpretability — not a trade-off between performance and transparency, but both.
Graph Neural Networks
GNNs applied to molecular graphs, learning directly from atom and bond connectivity — the most information-rich representation.
Large Language Models
LLMs applied to molecular SMILES strings, capturing chemical language patterns learned from large-scale training data. Used for a subset of endpoints where sequence-based representations yield strong predictive performance.
Consensus Modelling
Predictions combined across SMILES, fingerprints, graphs, and descriptors — reducing noise, bias, and variance of any single model.
OECD-Aligned Validation
Every model follows OECD Principles 1–5: defined endpoint, unambiguous algorithm, applicability domain, goodness-of-fit metrics, and mechanistic interpretation.
Mechanistic Transparency
Structural alerts and read-across methods ensure users understand the evidence behind each prediction — essential for regulatory confidence.
Continuous Cloud Updates
Models are updated as new training data becomes available. Your subscription always accesses the latest, most accurate versions.
See it in action with your own compounds.
Free trial — 2 compounds, no credit card required.