Structure & drug discovery
QSAR & Cheminformatics Modelling
We build ligand-based models that predict activity or properties from chemical structure, validated with cluster-aware cross-validation, Y-randomisation and an explicit applicability domain, and documented in a model card.
Who it is for
- Medicinal-chemistry groups
- Teams screening libraries before purchase or synthesis
- Projects that need a defensible in-silico property model
What you send us
- Measured activitiesSMILES with values and assay descriptions.
- Or a targetFor curation of public data from ChEMBL.
- Compounds to predictYour library or candidate list.
What you receive
- Curated dataset
- Validated model and performance report
- Predictions with applicability-domain flags
- Model card and code
Methodology
How the work runs
Each step is agreed with you and documented. Typical tools: RDKit · scikit-learn · XGBoost · ChEMBL · PubChem
- Data curationStructure standardisation, duplicate handling and harmonisation of activity types and units.
- Chemical spaceDescriptors, fingerprints, clustering and diversity analysis of the training data.
- ModellingRandom forest, kNN, gradient boosting or linear models, compared against simple baselines.
- ValidationCluster cross-validation, Y-randomisation and definition of the applicability domain.
- Prediction and reportPredictions with domain flags, a model card and the code to re-run them.
Quality controls
Checks applied to every project
- Duplicates and conflicting measurements resolved and documented
- Performance reported on cluster-based splits, not only random splits
- Y-randomisation to rule out chance correlation
- Every prediction flagged as inside or outside the applicability domain
Scope and limits
What the work can and cannot show
- A model is only as reliable as its training data; small or noisy datasets give wide error bars, which we report.
- Predictions outside the applicability domain are flagged rather than trusted.
- Research use only: results are not a clinical or diagnostic report.
Free quote
Start a QSAR project.
Describe your data and questions. We reply with questions or a written quote.
It depends on the data type, the number of samples and the depth of interpretation you need. Every project receives a written quote with scope, deliverables, price and delivery date after a short exchange about your data and questions. The quote is free and carries no obligation.
Data are used only for your project, transferred through a channel you approve and deleted at the end of the engagement on request. We will sign your institution’s confidentiality agreement before any data are shared.
Yes. You receive a methods section with software versions and parameters, and every figure can be regenerated from the delivered code. Authorship or acknowledgement follows your field’s norms and is agreed at the start.
Related