Utilizing SciPy and other open source packages to provide a powerful API for materials manipulation in the Schrödinger Materials Suite

TL;DR

The paper addresses the challenge of efficiently exploring vast materials design spaces by integrating open-source software with the Schrödinger Materials Science Suite to streamline structure generation, high-throughput workflows, and property prediction. It presents concrete workflows that couple open-source tools (OpenBabel, RDKit, pymatgen, Enumlib, SciPy, lmfit, matminer, DScribe, RDKit, DeepChem, AutoQSAR) with proprietary engines to enable deposition/evaporation simulations, accurate EOS fitting, and machine learning–driven material discovery. Key contributions include scalable job orchestration across diverse queuing systems, deployment of convex-analysis and optimization techniques for mechanical and structural properties, and an active-learning framework that dramatically reduces the need for expensive DFT calculations while pursuing multi-property objectives. The work demonstrates that open-source integration accelerates discovery, improves reproducibility, and fosters community-driven enhancements within computational materials science.

Abstract

The use of several open source scientific packages in the Schrödinger Materials Science Suite will be discussed. A typical workflow for materials discovery will be described, discussing how open source packages have been incorporated at every stage. Some recent implementations of machine learning for materials discovery will be discussed, as well as how open source packages were leveraged to achieve results faster and more efficiently.

Figures (6)

• Figure 1: Example of a workflow for computational materials discovery.
• Figure 2: Some example products that compose the Schrödinger Materials Science Suite.
• Figure 3: Example of the job submission process.
• Figure 4: Left: The uniaxial stress/strain curve of a polymer calculated using Desmond through the stress strain workflow. The dark grey band indicates an inflection that marks the yield point. Right: Constant strain simulation with convex analysis indicates elongation at yield. The red curve shows simulated stress versus strain. The blue curve shows convex analysis.
• Figure 5: Active learning workflow for the design and discovery of novel optoelectronics molecules.