COMP6 Benchmark dataset for ML potentials
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.

COmprehensive Machine-learning Potential (COMP6) Benchmark Suite

This repository contains the COMP6 benchmark for evaluating the extensibility of machine-learning based molecular potentials.

If you use the COMP6 benchmark please cite this paper:

Active learning-based (ANI-1x):

Justin S. Smith, Ben Nebgen, Nicholas Lubbers, Olexandr Isayev, Adrian E. Roitberg. Less is more: sampling chemical space with active learning. The Journal of Chemical Physics 148, 241733 (2018), (


Please read the in the repository linked below for instructions on how to extract the COMP6 HDF5 (extention *.h5) files.

The following paper contains a description of the file format:

COMP6 Benchmark Results:

These results represent the errors (MAE/RMSE) over the entire benchmark using a single ML potential (column 1). Please read Section IID for a detailed description of the error metrics.

Please contact Justin S. Smith at if you'd like to add your results from the COMP6 benchmark.

Complete COMP6 benchmark results:

Potential Energy Relative Energy Force
ANI-1x1 1.93/3.37 1.85/2.95 3.09/5.29
ANI-11 5.01/16.9 3.01/6.97 3.70/7.13

Units: kcal/mol and kcal/mol/A (errors are NOT per atom) Error key: MAE/RMSE


Related work

ANAKIN-ME ML Potential Method:

Justin S. Smith, Olexandr Isayev, Adrian E. Roitberg. ANI-1: An extensible neural network potential with DFT accuracy at force field computational cost. Chemical Science, 2017, DOI: 10.1039/C6SC05720A

Original ANI-1 data:

Justin S. Smith, Olexandr Isayev, Adrian E. Roitberg. ANI-1, A data set of 20 million calculated off-equilibrium conformations for organic molecules. Scientific Data, 4, Article number: 170193, DOI: 10.1038/sdata.2017.193

Active learning and transfer learning-based (ANI-1ccx):

Justin S. Smith, Benjamin T. Nebgen, Roman Zubatyuk, Nicholas Lubbers, Christian Devereux, Kipton Barros, Sergei Tretiak, Olexandr Isayev, Adrian Roitberg. Outsmarting Quantum Chemistry Through Transfer Learning. ChemRxiv, 2018, DOI: []