Train a target rescorer

Learn one score for your target from the docking and co-folding signals of compounds you have measured (Vina, GNINA, SurfDock, DiffDock, Boltz-2 and MM-GBSA scores and PLIP contacts, plus ligand descriptors), and compare it with each single score on the same ligands to see which signals predict activity.
Estimated minimum cost: $—

Examples

Training data

Docking results for compounds you have measured, and their measured values.

One file per docked ligand, or one multi-ligand SDF whose titles name the ligands. Each ligand is matched to the labels by name: the SDF title, else the file name. Boltz-2 scores come from a chained Boltz-2 step or a features table.

AutoDock Vina poses

Vina affinity

or drop files here
GNINA poses

GNINA affinity and CNN score

or drop files here
SurfDock poses

SurfDock affinity

or drop files here
DiffDock poses

DiffDock confidence

or drop files here
MM-GBSA scores

ΔG from scores.csv

or drop files here
PLIP interactions

Contacts per residue and interaction type

or drop files here
Labels

One row per measured ligand: its id (matching the docking outputs), the measured value (pIC50, pKd, or 0 / 1 for inactive / active) and its SMILES.

or drop a file here

Model parameters

Choose the prediction task, the features added to the docking signals, the algorithms to compare and how they are validated.

Features
Algorithms