Scaling structure-based drug design with machine learning

Computer-aided drug design (CADD) techniques are a routine part of modern drug development. Performing modeling before experimental work begins maximizes the potential to discover a molecular structure with the best therapeutic potential among the vast number of possibilities. Quantum mechanical and classical mechanical calculations can predict a molecule’s physical properties, estimate its binding strength to a target protein, and generate unexpected structural diversity while maintaining a molecule’s biological properties. However, these calculations are often too slow to be applied at a large scale.

One way to speed calculations is to use a set of CADD data to train machine learning models to predict molecular features of drug candidates that are slow to compute directly. Machine learning extends the capability of 3D shape and electronic structure calculations in CADD workflows to potentially operate at the scale of billions of molecules. The combination of CADD and machine learning could even one day speed resource-intensive free energy perturbation (FEP) calculations, which predict binding affinity with accuracy comparable to experimental measurements.

Download white paper: Scaling structure-based drug design with machine learning