Representation of molecular structures with persistent homology for machine learning applications in chemistry

Machine learning (ML) accelerates first-principles screening for functionalized molecules and materials. Molecular representations are crucial for ML in chemistry, impacting performance. This paper introduces a concise molecular representation derived from persistent homology, an applied mathematics branch. Its applicability is demonstrated through high-throughput computational screening of the GDB-9 database (133,000+ organic molecules) to identify novel CO₂-selective interacting molecules. The chemically-driven persistence image representation is used to screen GDB-9, suggesting molecules/functional groups with enhanced properties.

Jacob Townsend, Cassie Putman Micucci, John H. Hymel, Vasileios Maroulas, Konstantinos D. Vogiatzis