#compchem #chemsky

AceForce 1.0 is available for non-profit use and demonstrations. Feel free to explore, experiment, and push its limits.

huggingface.co/Acellera/Ace...

AceForce 1.0 is just the beginning. Future iterations will bring even more accurate models, expanded datasets, and advanced optimizations. We’re excited to see how these advancements will shape computational drug discovery.

These advancements allowed us to run the entire JACS dataset for RBFE, except PTP1B (-2 charge). Using our QuantumBind-RBFE platform, AceForce 1.0's accuracy and correlation is generally better or similar than GAFF2.

With AceForce 1.0 we have overcome most of these issues:
- Extended atom element supported
- +1 and -1 charges allowed
- Runs at 2fs with similar accuracy

Nowadays, most NNPs have some key limitations for drug discovery experiments.
- Limited atom element support
- Only neutral molecules
- Slow throughput, restricted to 1fs runs

Our main goal of AceForce was to use it in RBFE calculations in an NNP/MM setting, where the internal energies of the ligand are governed by the NNP. In previous work, we showed how this approach improves accuracy: pubs.acs.org/doi/abs/10.1...

AceForce 1.0 is our first neural network potential (NNP) trained on millions of quantum mechanics data points. Initial benchmarks already show comparable accuracy against other relevant NNPs

Si te gustan los Zelda prueba el Tunic!