Lira: Rotational Invariant Shape and Electrostatic Descriptors for Small Molecules based on Real Spherical Harmonics

Background: Modern AI-based tools are increasing the number of protein structures available, creating an opportunity and a challenge for automated high-throughput drug discovery pipelines. The amount of data is overwhelming for current methods, thus demanding new high-performance approaches for Machine Learning-based rational drug design. Results: : As shape and electrostatics are the main elements for understanding protein-ligand interaction, we used them to create a system of efficient and AI-compatible descriptors. This method, called LIRA, describes, compares and analyses shape and electrostatics for small ligands. It can generate databases with descriptors for tens of millions of shapes in a few hours, which can then be queried in seconds. The Lira design, focused on performance and reliability, makes its integration into AI-driven rational drug design pipelines simple. Conclusions: : Given an input query, we demonstrate that spherical harmonics can be used as the basis of an accurate and efficient platform for searching, selecting and comparing molecules for shape and electrostatics. It can be employed to efficiently compare large numbers of molecules, establishing a robust and compact method for describing and comparing the molecular shape and electrostatic potential suitable for Machine and Deep Learning applications. It also can be integrated with other structural software to create signatures directly based on the pocket characteristics of protein targets, thus accelerating novel lead identification. In addition, ongoing medicinal chemistry campaigns can benefit from applying Lira as the starting point for AI-guided drug development and lead maturation.