2D QSAR modelling with proper validation
Turn a table of structures and measured activities into a predictive model. ReactorMD calculates descriptors, builds and validates QSAR models and reports the statistics that reviewers expect.
Who this service is for
- Researchers with experimental IC50, Ki or percent-inhibition data
- Students needing QSAR for a thesis chapter
- Groups wanting to predict activity for new analogues
What to send us
- Structures as SMILES or SDF with measured activity (for example pIC50)
- Activity units and assay type
- Optional external test set and the compounds you want predicted
How the work is done
- Curate the dataset: remove duplicates, check structures and convert activities to a consistent scale.
- Calculate 2D descriptors and fingerprints.
- Reduce descriptors by removing constant and highly correlated variables.
- Split into training and test sets.
- Build models and tune them.
- Validate with cross-validation, test-set statistics, y-randomisation and applicability domain.
- Predict new compounds and report the model.
What you receive
- Model equation or model file with chosen descriptors
- R², Q² and RMSE statistics with predicted versus observed plots
- Applicability domain analysis
- Predictions for new compounds
- Methods text and a ZIP with data and code outputs
2D QSAR modelling packages
| Dataset Size | Price (USD) | Estimated Turnaround |
|---|---|---|
| Up to 50 compounds | $1.99 | Up to 6 hours |
| Up to 100 compounds | $3.49 | Up to 6 hours |
| Up to 250 compounds | $6.99 | Up to 6 hours |
| Up to 500 compounds | $11.99 | Up to 12 hours |
| Up to 1,000 compounds | $21.99 | Up to 12 hours |
What this method can and cannot tell you
A QSAR model is only reliable for compounds similar to its training data. Small or noisy datasets give weak models, and the report states this clearly.
Frequently asked questions
How many compounds do I need for a QSAR model?
As a rough guide, at least 20 to 30 compounds with a reasonable activity range. More data and a wider range give more dependable models.
What is the difference between 2D and 3D QSAR?
2D QSAR uses descriptors calculated from the molecular graph. 3D QSAR uses aligned 3D shapes and fields. 2D models are faster and need no alignment.
Which validation metrics do you report?
R², cross-validated Q², RMSE, external test-set performance, y-randomisation and applicability domain.
Can you predict activity for my new compounds?
Yes. Provide the structures and they are predicted with a note on whether they fall inside the model's applicability domain.
Related services
In silico ADME and drug-likeness prediction: Lipinski and Veber rules, physicochemical descriptors, solubility, permeability and toxicity alerts for compound prioritisation.
ServiceVirtual screeningStructure-based virtual screening of compound libraries: library preparation, high-throughput docking, filtering and a ranked, traceable hit list.
ServiceMolecular dockingOnline protein–ligand molecular docking: structure and ligand preparation, binding-site definition, pose ranking, interaction analysis and a full report.
Learn more
Ready to start?
Tell us about your target and compounds. The price and scope are confirmed before payment.
Computational results are predictions and support, but do not replace, experimental validation. Page reviewed by ReactorMD, 16 October 2026.