SandboxAQ introduced SAIR (Structurally Augmented IC50 Repository), a groundbreaking open dataset set to transform computational drug discovery. This vast and highly detailed collection, featuring millions of protein-ligand pairs with experimental potency data, offers an unprecedented resource for researchers. It's designed to empower advanced AI models, drastically improving the speed and accuracy of binding affinity predictions. Developed with support from NVIDIA DGX™ Cloud and leveraging SandboxAQ's advanced Large Quantitative Model (LQM) capabilities, SAIR includes approximately 5.2 million synthetic 3D molecular structures across over a million protein-ligand systems. This collaboration also boosted GPU utilization by 2x, optimizing SandboxAQ's scientific workflows. SAIR's unique power comes from its integration of physics-based modeling with LQM, providing enhanced reliability, improved generalization, and broader applicability for various drug discovery tasks.
SAIR was created to achieve previously impossible accurate, large-scale in silico predictions of protein-ligand binding affinities. This initiative fundamentally transforms the traditional, often slow, trial-and-error approach into a rapid, data-driven process. By making over five million affinity-labeled protein-ligand structures publicly available, SandboxAQ provides scientists with the essential "raw fuel" to quickly train breakthrough models, accelerating the pace of drug discovery.
The SAIR dataset enables AI models to deliver predictions at least 1,000 times faster than conventional physics-based methods, significantly accelerating the journey of new drugs from discovery to market. SandboxAQ's quantitative AI technology is already demonstrating superior outcomes through partnerships with leading institutions like UCSF’s Institute of Neurodegenerative Diseases and pharmaceutical innovators, consistently achieving higher hit rates compared to traditional methods. Researchers can access the SAIR dataset on Google Cloud Platform or directly from the SandboxAQ website.