Ready for Universal Quantum Chemistry? 🔬🧪👩‍🔬

So essentially,

We have a Universal Quantum Chemistry dataset to find new drugs now!

Paper: ∇2DFT: A Universal Quantum Chemistry Dataset of Drug-Like Molecules and a Benchmark for Neural Network Potentials (18 Pages)

Github: https://github.com/AIRI-Institute/nablaDFT

Researchers from Moscow and St. Petersburg want to make better methods of computational quantum chemistry provide accurate approximations of molecular properties. This is crucial for industries such as computer-aided drug discovery and other areas of chemical science. However, high computational complexity limits the scalability of their applications.

Source: https://lexica.art/prompt/c1d758e1-d7fe-4a56-9d01-7bebfb643427

Hmm..What’s the background?

Quantum chemistry is used to approximate molecular properties which are important for areas such as computer-aided drug discovery. The complexity of these computations makes them difficult to scale. Density functional theory (DFT) is the primary approach for solving the many-particle SE for electrons. DFT provides reasonably accurate predictions but even a single iteration may take several CPU hours making its use in molecular modeling tasks where it would be called many times, impractical.

Neural Network Potentials (NNPs) are a promising alternative to quantum chemistry methods but require large datasets for training. The researchers decided to release a dataset to assist NNP approaches.

Ok, So what is proposed in the research paper?

The researchers created a new dataset and benchmark called ∇2DFT. The ∇2DFT dataset is built on the nablaDFT dataset, contains twice as many molecular structures, and three times more conformations. The data was generated using the ωB97X-D/def2-SVP DFT level.

They proposed a benchmark encompassing Hamiltonian prediction, energy and force prediction, and conformational optimization to evaluate the performance of NN-based models for QC.

Source: Paper

What’s next?

Future work may involve expanding the ∇2DFT dataset to include structures not currently included such as nanoparticles, nanotubes, big rings, and other non-drug-like structures.

The authors plan to release a full set of interatomic forces for the ∇2DFT dataset, training NNPs on the full ∇2DFT, benchmark other NNPs for conformational optimization using the ∇2DFT dataset.

So essentially,

We have a Universal Quantum Chemistry dataset to find new drugs now!