Nvidia-backed Sandboxaq accelerates drug discovery by launching AI molecular datasets

Sandboxaq, a Nvidia-backed AI startup spun from Alphabet’s Google, has released a large dataset that hopes to speed up drug development. the goal? It helps scientists understand more quickly whether drugs stick to the protein they are targeting.

Drug protein binding is a fundamental part of drug discovery. If treatment does not properly bind to targets in the body, it is unlikely to function. However, understanding this usually involves months of lab testing or heavy calculations.

That’s where Sandboxaq comes in. Instead of relying on physical experiments, the startup used NVIDIA chips to generate data based on real scientific results. Although these molecules do not exist in nature, they were created using equations that reflect how grounded atoms interact with experimental data.

The company has its dataset publicly and hopes others can use it to train AI models that can predict drug-to-drug binding with high accuracy, but only in a fraction of the time. Sandboxaq is billed for access to its own trained model. I believe this can be done on par with some lab-based methods.

“This is a long-standing biology problem and we are all trying to solve it as an industry,” Nadia Harhen, GM of AI Simulations at Sandboxaq, told Reuters. “All of these computationally generated structures are tagged with true experimental data on earth, so if you select this dataset and train your model, you can actually use the synthetic data in a way that has never been done before.”

For researchers, datasets can be shortcuts. If you are testing whether a new drug can destroy biological processes, that is, slow the progression of the disease, this tool can help you predict whether the drug will actually bind to the appropriate target without having to run an experiment first.

This effort is part of a growing trend to tackle harsh scientific problems by combining physics-based models with machine learning. Although traditional methods can model atomic behavior, the number of possible combinations is overwhelming, even for small pharmaceutical compounds. Sandboxaq’s AI training model aims to simplify it.

The startup has raised over $950 million since it spun out of the alphabet in 2022. The latest $450 million Series E rounds include Raydario, Horizon Kinetics, BNP Paribas, Google, Nvidia and T. Support was provided by Low Price, Blayer Capital and others. The company says the funding will help expand work across multiple industries, including biotechnology, cybersecurity, financial services and materials science.

This release shows a huge drive to biopharma. By opening the dataset, Sandboxaq bets that the next big breakthrough in drug discovery could come from the keyboard rather than from the test tube.

Last week, we introduced Sandboxaq in our series.