SandboxAQ, an artificial intelligence startup that originated from Alphabet’s Google and is backed by Nvidia, announced the release of a large dataset on Wednesday aimed at
accelerating the development of new medical treatments. The dataset is designed to help researchers better understand how drugs bind to proteins, a critical factor in drug discovery.
The company’s goal is to assist scientists in predicting whether a drug will effectively attach to its target in the human body. Although the data aligns with real scientific results, it was not collected in a laboratory. Instead, SandboxAQ used Nvidia’s high-performance chips to generate the data computationally. This synthetic data will now be used to train AI models that can quickly determine whether a small-molecule drug is likely to bind to a specific protein involved in a disease or biological process.
This method falls within a growing field that merges classical scientific computing with modern AI techniques. Scientists already have equations to describe how atoms combine into molecules, but predicting interactions between complex, three-dimensional pharmaceutical compounds and proteins involves a huge number of possible combinations. These are too vast to calculate manually, even with advanced computers.
To address this, SandboxAQ created approximately 5.2 million synthetic molecular structures using real experimental data and mathematical models. These molecules do not exist in nature but are grounded in accurate scientific principles. The company is releasing this synthetic dataset publicly so that researchers can train AI systems to make predictions much faster and with high accuracy.
While the data is free, SandboxAQ plans to offer proprietary AI models built on the dataset as commercial tools. The goal is to achieve results that are as reliable as lab experiments, but far quicker and entirely virtual.
"This is a long-standing challenge in biology that the entire industry has been working to solve," said Nadia Harhen, general manager of AI simulation at SandboxAQ, in an interview with Reuters. "Each of these computer-generated structures is linked to verified experimental data, making it possible to use the synthetic dataset in ways that have not been done before."
Found this article interesting? Follow us on X(Twitter) ,Threads and FaceBook to read more exclusive content we post.
