QSAR Model Studio | BayesPharma Innovator

Build, validate, predict
chemical activity models

A professional QSAR/QSPR workflow for chemical activity prediction: validate SMILES data, generate molecular descriptors, train models, compare algorithms, assess applicability domain, run external validation, and create downloadable model reports.

Powered by Uni-Mol · Graphormer · ChemBERTa · DeepChem

How to create a QSAR model report

1. Add training dataPaste SMILES and observed activity values in JSON format.

2. Validate datasetCheck invalid SMILES, missing activity, duplicates, and activity range.

3. Build modelTrain Random Forest, Extra Trees, Ridge, PLS, MLP, XGBoost, Uni-Mol, Graphormer, ChemBERTa, or DeepChem.

4. Predict & reportPredict new compounds, check domain, and download/share the report.

Use Create Complete QSAR Report for the full workflow: model build, training prediction, y-randomization when possible, algorithm comparison, model card, quality checklist, and downloadable report package.

AI-powered models from Microsoft, Meta, IBM, and DeepChem included

Scientific guardrails

QSAR predictions are only as reliable as the dataset. Small datasets are useful for prototypes, not final decision-grade models. Use external validation, applicability-domain checks, assay harmonization, and expert review.

Check SMILES Remove duplicates Confirm units External validation Domain check

Report outputs

View full report Download JSON Download HTML Download Word Download CSV Share summary

Training Dataset

Activity field

Default algorithm

Active model ID

Report title

Training data JSON

Build Model

AI Models

Build a regression model from the training dataset. The returned model ID is automatically saved in the Active Model ID field.

Uni-Mol: 3D molecular representation (Microsoft) Graphormer: Graph Transformer (Microsoft) ChemBERTa: SMILES language model (IBM) DeepChem: Ensemble toolkit

Algorithms

Predict New Compounds

Prediction compounds JSON

Prediction Summary

After prediction, this panel will show predictions, inside/outside domain status, and confidence flags.

External Validation

Use compounds with observed activity values to evaluate an existing model.

External validation data JSON

Quality Checklist

SMILES validity Activity numeric No duplicates Train/test metrics Cross-validation Applicability domain External validation Y-randomization

For strong scientific credibility, do not rely only on training performance. Use external validation and domain checks.

Model Registry

Model Card

Complete Report Center

Create a complete report, then view it here or download/share it.

Raw JSON

{}