AI BriefingSAIR Releases 5.24 Million-Entry Dataset
Key point
SandboxAQ has released SAIR, a dataset of 5.24 million 3D protein-ligand structures for predicting drug efficacy.
Details
SandboxAQ has launched SAIR (Structurally Augmented IC50 Repository). This dataset contains 5.24 million AI-generated 3D protein-ligand structures combined with experimentally measured IC₅₀ (half maximal inhibitory concentration) labels, enabling a direct link between molecular structure and drug efficacy.
Existing AI-based drug design has faced limitations in predicting efficacy or toxicity from molecular structure alone, but SAIR bridges this gap by providing high-quality 3D structures paired with validated binding efficacy data at scale. This allows the hit-to-lead discovery and optimization process to shift from the wet lab to in silico simulation, accelerating R&D speed.
The dataset was built using the Boltz1 model and 760 NVIDIA H100 processors, requiring a total of 130,000 GPU hours. Through optimization using NVIDIA DGX Cloud, data generation was completed 4 times faster than originally expected.
SAIR is released under a CC BY 4.0 license, making it freely available for both commercial and non-commercial research purposes.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Stanford and NVIDIA Release 'Contrastive Language Model' CLM with 9x Faster Inference