Numerion Labs Publishes Breakthrough AI Research on Ultra-Fast Screening of Hyper-Scalable Virtual Libraries with NVIDIA

APEX Protocol Enables Exhaustive Evaluation of 10 Billion Virtual Compounds in Under 30 Seconds

SAN FRANCISCO, Oct. 29, 2025 /PRNewswire/ -- Numerion Labs, an AI-native company pioneering the application of AI/ML to drug discovery, today announced the release of a research paper co-authored by NVIDIA subject matter experts, now available on arXiv. The paper introduces APEX (Approximate-but-Exhaustive Search), a new computational protocol that allows scientists to screen hyper-scalable, ultra-large combinatorial synthesis libraries (CSLs) containing billions of compounds in seconds, dramatically accelerating workflows that previously took months to analyze.

Numerion Labs, formerly known as Atomwise, developed APEX to address one of the most significant bottlenecks in virtual screening: the inability to comprehensively search chemical space at scale. By pairing deep learning surrogates with GPU-accelerated enumeration over structured chemical spaces, APEX virtually evaluates billions of potential starting points in seconds, ensuring that promising compounds are not overlooked.

"Drug discovery has always faced a fundamental challenge of searching chemical space at vast scale and speed," said Steve Worland, CEO of Numerion Labs. "With APEX, we've demonstrated that it is now possible to virtually evaluate billions of molecules in seconds, enabling a real-time marriage of creativity between chemist and computer. This ability to rapidly explore a far broader chemical space yields a much more diverse set of chemical starting points, accelerating discovery and increasing the likelihood of identifying differentiated new medicines to treat disease."

Traditional virtual screening techniques typically assess less than 0.1% of available compounds, leaving many valuable potential drugs undiscovered. Competing ultra-large library tools often focus on compound similarity or physicochemical filters – or rely on brute-force docking methods – that require massive computational resources. In contrast, APEX leverages COSMOS – Numerion's structure-based, generative pre-trained foundation model – to prioritize compounds with true biological relevance. Trained to predict molecular binding and function, COSMOS enables APEX to go beyond identifying molecules that merely appear drug-like, and instead selects those most likely to bind disease-driving targets with optimal drug-like properties. In benchmark tests, APEX retrieved the top one million biologically promising compounds from a 10-billion-compound library in under 30 seconds on a single NVIDIA GPU.

The research demonstrates APEX's ability to identify high-scoring compounds that meet key drug-like property constraints across a diverse range of protein targets, including kinases, GPCRs, proteases and nuclear receptors. These advances have the potential to fundamentally reshape how pharmaceutical and biotechnology companies approach hit discovery by enabling comprehensive, real-time exploration of chemical space. This, in turn, can accelerate progression to clinical candidates and increase the probability of success in discovering differentiated new medicines.

The full paper is available on arXiv here, and the code is available on GitHub here.

About Numerion Labs
Numerion Labs is an AI-native company accelerating the discovery of life-saving medicines through the development and use of cutting-edge machine learning algorithms. The company unites computational chemistry, structural biology, and medicinal chemistry to pioneer the next generation of AI-driven drug discovery platforms.

Media Contact:

Kate Thompson / Tim Ragones / Madeline Jones
Joele Frank, Wilkinson Brimmer Katcher
212-355-4449

SOURCE Numerion Labs