Robust and efficient reranking in crystal structure prediction: a data driven method for real-life molecules

Abstract

We accelerate a key step in crystal structure prediction (CSP) using machine learning and report its robustness on a wide array of pharmaceutical molecules. The speedup achieved by our scheme allows for a scale-up in both the number of candidate drug molecules studied and the level of theory employed in their treatment, paving the way for tackling more complex crystal energy landscapes.