SandboxAQ announced today a groundbreaking advance that pushes the boundaries of computational chemistry, impacting biopharmaceuticals, chemicals, materials science and other industries. Collaboration with NvidiaSandboxAQ leverages immense quantitative models (LQM) and the NVIDIA CUDA-accelerated Density Matrix Renormalization Group (DMRG) algorithm to enable researchers to perform highly exact quantitative AI simulations of real-world systems with demanding accuracy, going beyond what immense language models (LLM) and other AI models can currently do.
The combination of the CUDA-DMRG algorithm, the NVIDIA Quantum platform, and NVIDIA accelerated computing speeds up these highly exact calculations by more than 80x compared to classic calculations on a 128-core processor. At the same time, it doubles the size of the computable catalysts and enzyme lively sites computed by the system. SandboxAQ researchers exploit these computational results to train AI networks to optimize for a desired treatment or catalyst, as described in an available preprint HERE.
Last year, SandboxAQ announced Collaboration in Artificial Intelligence from University of California San Francisco (UCSF), Novonix, and Riboscience. In 2024, Flagship Pioneering, SPARK NS, and other organizations signed an agreement to expand their innovation channels.
Applications of LQM include biopharma, agriculture, and advanced materials. For example, in biopharma, the enzyme Cytochrome P450 plays a central role in human drug metabolism and is crucial to understanding drug toxicity. CUDA-DMRG can aid solve the long-standing problem of accurately modeling cytochrome catalytic activity and provide a breakthrough angle for computational toxicity prediction, enabling computational simulation to reduce the risk of clinical trials before they even begin.
Training immense AI models with proprietary, generated data to unlock breakthroughs in the physical world is at the heart of the recent wave of quantitative AI. LQMs can make exact predictions about the world because they rely on exact, physics-based data. While LLMs are confined to data available on the Internet or other existing sources, SandboxAQ LQMs can access an unlimited pool of training data generated by physics-based quantitative AI simulations.