Custom QSAR/QSPR Models

Train predictive models
on your data

Build quantitative structure-activity and structure-property relationships tuned to your specific chemical space. Train models on your assay, your catalysts, your materials, your molecules.

Parity PlotR² = 0.941
Train Test

Custom model training & validation

Train models directly on your assay results, reaction outcomes, or property measurements. Choose from diverse models, hyperparameters, and validation approaches.

Feature analysis & selection

Run correlation analysis to find and remove redundant features, then refine with recursive feature elimination to surface features that matter.

Deploy your models to make predictions

Compare trained models and deploy them to score new candidate structures or experiments.

Mechanistic insights

Gain actionable chemical knowledge by interrogating models with feature importance calculations.

See it in action

From data to predictions in four steps

Upload your data, train models, validate performance, and predict on new molecules — all from the browser.

1

Upload

Load your training data

Upload a CSV with molecular descriptors (or SMILES) and your measured property. The platform auto-detects columns, flags missing values, and lets you pick the target variable.

Training Data Preview

7 columns · 64 rows in loaded preview

Target YFeature XExcluded
rowsmilesMaxAbsEStateMaxEStateMinAbsEStateqedSPSpIC50
1O1CC(C@@H)N…0.7310.7310.0880.6190.274
2Fc1cc(c(F)(C…0.8670.8670.0290.2190.399
3S1(=O)(=O)N…0.8160.8160.0140.3650.177

Y pIC50·X 187 features·Excluded 6

Validated
2

Analyze

Surface the features that matter

Compute Spearman correlations against your target, flag redundant descriptor pairs, and rank features by permutation importance.

Target Correlation
Spearman ρ vs pIC50
MaxAbsEState
+0.82
qed
+0.74
TPSA
+0.69
MolLogP
+0.61
NumHAcceptors
+0.53
FractionCSP3
-0.47
Permutation Importance
Top 4
MaxAbsEState
qed
TPSA
MolLogP

187 features analyzed · 6 highly correlated pairs flagged

12 selected →
3

Train

Train, validate, and compare

Train models directly on your assay results, reaction outcomes, or property measurements. Choose from diverse models, hyperparameters, and validation approaches.

Model Training
FINISHED

Model

Random Forest

Random Forest
XGBoost
ElasticNet
Bayesian Ridge

Features

Auto · 10

Split

K-Fold (5)

HP Tuning

Auto Tune

Parity PlotR² = 0.63
Train Test
Fold Metrics
MetricBest TrainBest TestAvg Test
0.810.680.63
Adj-R²0.770.13-0.00
MSE0.330.620.64
MAE0.440.620.62
Model Comparison — Avg Test R²
Random Forest
0.63
XGBoost
0.88
ElasticNet
0.41
Bayesian Ridge
0.52
Random ForestXGBoostElasticNetBayesian Ridge1 of 4 trained
4

Predict

Score new compounds instantly

Upload inference file with candidate structures. The platform performs inference with your trained model and computes predicted properties for your unseen molecules.

Applications

Predict before you synthesize

Drug Discovery & ADMET

Predict pIC50, EC50, solubility, permeability, metabolic stability, and other biological or ADMET endpoints. Prioritize which candidates to synthesize and flag compounds with unfavorable properties early.

Materials Property Prediction

Predict Tg, tensile strength, HOMO-LUMO gap, and other target properties from molecular or polymer descriptors. Design novel materials with ML-guided optimization of synthesis conditions.

Reaction Outcome Prediction

Train on reaction descriptors and experimental conditions to predict yield, selectivity, and more. Identify which molecules to test next before going to the bench.

Ready to build your first model?

Upload your data, train a custom model, and start predicting — all in minutes.