Genesis Molecular AI announces potency prediction breakthrough with Ruby, releases potency training dataset

News originally published by

GlobeNewswire

Ruby led all evaluated methods on four public potency benchmarks and achieved unprecedented accuracy on real drug programs from the Genesis-Incyte collaboration

SAN MATEO, Calif., Nov. 20, 2026 (GLOBE NEWSWIRE) -- Genesis Molecular AI, the frontier AI lab for drug discovery, today published a technical report on Ruby, a model that achieves a step change in performance in predicting small-molecule potency. Ruby benefits from structural representations learned by Pearl, the company’s state-of-the-art foundation model for cofolding, as well as public and proprietary potency assay training data. Genesis is also releasing GenesisPotency-2026, an audited public-source portion of Ruby’s training data, under a Creative Commons Attribution-ShareAlike 4.0 license.

Ruby is the only model evaluated that performs well at both ends of the discovery process: screening large libraries with high enrichment, and ranking closely related molecules within a chemical series. It is faster and more accurate than the other methods evaluated, for a few reasons.

Ruby is built on Pearl, Genesis’s state-of-the-art protein–ligand structure prediction model announced at NVIDIA GTC DC in 2025 (see report jointly authored by Genesis and NVIDIA). Ruby uses Pearl’s internal structural representations directly rather than needing to generate a 3D pose for every compound. Removing that as a required step cuts the dominant computational cost. The resulting throughput is increased further by architectural and compute optimizations developed using NVIDIA libraries, which are acknowledged in the technical report. The curated training data described below improves prediction accuracy on top of that.

Key findings from the report:

  • Ruby achieved the highest average correlation among all methods evaluated on four public potency benchmarks, a field that included Isomorphic Labs’ IsoDDE and physics-based free-energy methods.
  • Across eight assays from five Genesis-Incyte collaboration programs, Ruby ranked compounds substantially more accurately than all other methods tested, both before and after incorporating earlier lab results, and had the lowest average prediction error after calibration.
  • At its fastest setting, Ruby scored compounds in 25 milliseconds each, between roughly 40 and more than 1,000 times faster per compound than the other methods evaluated, while ranking known actives at least as effectively.

“Chemists don’t adopt a model because it wins a benchmark. They adopt it when it holds up on their own program, on chemistry or targets that nobody has seen before,” said Evan Feinberg, Ph.D., founder and CEO of Genesis. “That’s why these results on our Incyte collaboration programs matter most to us. Ruby is performant on those datasets even before seeing a single measurement from them.”

An open dataset for the field

Genesis is also releasing GenesisPotency-2026, a curated training set assembled from the public potency literature, under a Creative Commons Attribution-ShareAlike 4.0 license.

Potency measurements are difficult to work with across public sources. Records are inconsistently annotated, and because databases often share the same underlying source documents, agreement between them can reflect a shared error rather than independent confirmation. Genesis went back to the source documents themselves, checking target identity, experimental conditions, endpoints and units, and corrected or set aside records that did not hold up. In one case, 885 measurements of inhibitors of one protein had been assigned to an unrelated protein in two separate databases.

The result is a single consolidated dataset released with row-level provenance and the review annotations attached, so researchers can trace any measurement back to its source, see how it was assessed, and apply different standards if they disagree.

About Genesis Molecular AI

Genesis Molecular AI, Inc. is the frontier AI lab for drug discovery, building foundation models to unlock a new era of agentic drug design and development. The company’s generative and predictive AI platform, GEMS (Genesis Exploration of Molecular Space), unifies AI and physics into industry-leading models to generate and optimize drug molecules, including the Pearl structure prediction model and the Ruby potency prediction model. Genesis is backed by premier AI and life science investors, including a16z, NVIDIA, Rock Springs Capital, Menlo Ventures, T. Rowe Price, Fidelity, and Radical Ventures.

Media contact
press@genesis.ml


Primary Logo

Legal Disclaimer:

EIN Presswire provides this news content "as is" without warranty of any kind. We do not accept any responsibility or liability for the accuracy, content, images, videos, licenses, completeness, legality, or reliability of the information contained in this article. If you have any complaints or copyright issues related to this article, kindly contact the author above.

Share this page:

Advanced Search Options

Search for:

Search scope:

Type:

Search in:

Date range:

The last

Sort by:

Sign up for:

Sci-Tech North Carolina

The daily local news briefing you can trust. Every day. Subscribe now.

By signing up, you agree to our Terms & Conditions.