PreditX 2.0 – One-Click Model Setup

Submit once. We'll prepare data, engineer features, select variables, and train your chosen models in the background. You'll get an email with a unique link to Step 6: Molecule Prediction.

Submit Job

Pulls target-specific bioactivities from ChEMBL, standardizes units (nM → pIC50), removes salts/duplicates, and harmonizes assay types. Outputs: *_bioactivity_data_labeled.xlsx and QC plots (activity distributions).

Computes pIC50, applies activity thresholds (Active / Intermediate / Inactive), and prepares the dataset for modeling. Outputs: labeled table + histograms.

Generates Bemis–Murcko scaffolds, performs scaffold-aware train/test split, and computes PCA/MDS and Tanimoto heatmaps. Outputs: *_training_set.xlsx, *_test_set.xlsx, diversity plots.

Calculates Mordred descriptors and/or Morgan fingerprints, removes constant/correlated features, optional PCA, and stores selected features for inference. Outputs: feature matrices, PCA objects, feature lists.

Trains chosen algorithms with tuning and cross-validation, saves best estimators, confusion matrices, ROC curves, and performance tables. Outputs: .pkl models, model_performance_tuned_*.xlsx, plots.