Introduction
Atomwise is a San Francisco biotechnology company founded in 2012 that pioneered convolutional neural network approaches to structure-based drug discovery — and, as of its rebrand to Numerion Labs, operates as an AI-native organization uniting computational chemistry, structural biology, and medicinal chemistry under one discovery platform. Abraham Heifets co-founded Atomwise; Steve Worland, Ph.D., serves as chief executive officer of Numerion Labs per the team page reviewed in July 2026. The atomwise.com domain redirects to numerionlabs.ai, reflecting the corporate identity transition Brel notes in import metadata.
On October 29, 2025, Numerion Labs announced APEX (Approximate-but-Exhaustive Search), a computational protocol co-authored with NVIDIA subject matter experts that screens ten-billion-compound combinatorial libraries in under thirty seconds on a single NVIDIA GPU — a throughput claim Worland tied to real-time collaboration between chemists and computers exploring vastly broader chemical space.
What the company does
Numerion Labs develops machine learning algorithms and computational protocols that accelerate hit discovery from virtual compound libraries orders of magnitude larger than traditional docking workflows can exhaustively search. COSMOS, the company’s structure-based generative pre-trained foundation model, predicts molecular binding and function to prioritize biologically relevant compounds rather than molecules that merely appear drug-like by physicochemical filters alone.
APEX pairs deep learning surrogates with GPU-accelerated enumeration over structured chemical spaces, enabling scientists to virtually evaluate billions of potential starting points in seconds. Worland stated in the October 2025 release that traditional virtual screening assesses less than zero point one percent of available compounds — leaving valuable drug candidates undiscovered — while competing ultra-large library tools rely on brute-force docking requiring massive compute clusters.
Numerion Labs sells discovery capabilities to pharmaceutical and biotechnology partners while advancing internal pipeline programs — a hybrid model common among AI drug discovery companies transitioning from software licensing toward owned therapeutic assets.
Who it serves
Numerion Labs serves pharmaceutical and biotechnology R&D organizations running hit discovery against diverse target classes including kinases, GPCRs, proteases, and nuclear receptors — protein families APEX benchmark tests covered in the October 2025 research announcement. Medicinal chemistry teams evaluate APEX on whether retrieved top-scoring compounds meet drug-like property constraints and represent genuinely novel chemical matter rather than analogs of known scaffolds.
Discovery groups constrained by months-long virtual screening cycles represent Numerion’s core buyer — teams where exhaustive library search was computationally impossible before GPU-accelerated surrogate models compressed timelines to seconds.
Company background
Heifets co-founded Atomwise in 2012 in San Francisco, building among the earliest deep learning platforms for molecular binding prediction. The company partnered with major pharmaceutical organizations and advanced multiple discovery collaborations before rebranding as Numerion Labs to reflect an AI-native identity spanning foundation models, structural biology, and medicinal chemistry. Worland leads the company as CEO through the APEX and COSMOS platform generation, per Numerion’s team page.
The Atomwise-to-Numerion transition mirrors broader AI biotech rebrands as companies outgrow original algorithm-only positioning and pursue integrated discovery platforms with published research validating throughput claims.
Product and AI capabilities
APEX leverages COSMOS to prioritize compounds with true biological relevance — predicting binding and function rather than similarity to known drugs. Benchmark tests described in Numerion’s October 2025 release retrieved the top one million biologically promising compounds from a ten-billion-compound library in under thirty seconds on one NVIDIA GPU. The research paper was released on arXiv with code on GitHub, per the official announcement — enabling external validation of performance claims.
Worland framed APEX as enabling a real-time marriage of creativity between chemist and computer — a workflow shift from batch overnight docking jobs toward interactive exploration during live discovery sessions.
Key developments
2012: Atomwise founded in San Francisco by Abraham Heifets. October 29, 2025: Numerion Labs published APEX protocol research with NVIDIA (numerionlabs.ai press release, official source). 2026: Steve Worland, Ph.D., CEO per Numerion Labs team page; atomwise.com redirects to numerionlabs.ai.
Numerion’s APEX announcement explicitly states the company was formerly known as Atomwise — Brel retains Atomwise as profile name with rebrand noted in import metadata for search continuity.
Why it matters
Atomwise — now Numerion Labs — matters because virtual screening throughput determines how much chemical space discovery teams can explore before committing wet-lab synthesis budgets. APEX’s published benchmark of ten billion compounds in thirty seconds is a concrete 2025 milestone, not a roadmap claim, co-authored with NVIDIA and released with open code. Worland’s emphasis on diverse chemical starting points addresses a known failure mode of narrow screening libraries that produce undifferentiated clinical candidates.
Numerion sits in Brel biotechnology coverage near Deep Genomics and Healx but focuses on small-molecule hit discovery rather than RNA therapeutics or rare-disease repurposing — buyers should match platform chemistry to modality strategy.
Sector context
Atomwise belongs in Brel’s AI in healthcare coverage under biotechnology and AI drug discovery. Readers evaluating Numerion Labs should review APEX benchmark methodology on arXiv and distinguish published research throughput from partner pipeline clinical outcomes — hit discovery acceleration does not guarantee development success.
Pharma discovery leaders comparing AI screening vendors should ask for target-class-specific validation data — kinase benchmark performance does not automatically transfer to GPCR programs with different binding site physics.
Sources and references
Numerion Labs APEX research press release (October 29, 2025) and Team leadership page.
- Numerion Labs — APEX research with NVIDIA (numerionlabs.ai, October 29, 2025)
- Numerion Labs — Team / leadership (numerionlabs.ai)
Reviewed July 10, 2026. Rebranded from Atomwise to Numerion Labs. Steve Worland is CEO; Abraham Heifets is co-founder.