Genesis Molecular AI Unveils Pearl, a Field-Leading Foundation Model that Achieves Unprecedented Performance in Drug-Protein Structure Prediction

Audio By Carbonatix

BURLINGAME, Calif.--(BUSINESS WIRE)--Oct 28, 2025--

Genesis Molecular AI, the company pioneering the world’s leading molecular AI models for drug design and development, today announced Pearl, a generative foundation model for biomolecular structure prediction trained with novel architecture, training methodology, and large-scale synthetic data. Genesis released a new study, co-authored with subject matter experts from NVIDIA, demonstrating that Pearl outperforms existing models, including AlphaFold 3, in successfully predicting how small molecules bind to proteins.

Drug-protein structure prediction is considered a “holy grail” in drug discovery. Solving this historically difficult problem holds enormous promise for designing the next generation of medicines for patients with severe unmet medical needs. While popular large language models had access to vast quantities of readily accessible training data from the internet, AI for biochemistry has no such luxury.

Pearl is an end-to-end diffusion model that substantially improves over existing models by advancing Generative AI techniques in low data regimes. Pearl uniquely exploits physics in its input, architecture, and outputs. A key innovation behind Pearl's field-leading capabilities is training on large scale synthetic data from simulation, to overcome the scarcity of high quality experimental structural data. Pearl’s performance improves consistently as more simulated data is added, establishing the first evidence of synthetic data scaling laws in AI-driven drug discovery. This breakthrough crucially enhances the model’s ability to generalize beyond limited public data.

"One of the biggest roadblocks in applying AI to drug discovery is the lack of high-quality biomolecular data, which are expensive and time-consuming to obtain. Autonomous vehicle models surmounted analogous challenges with simulated data, which inspired Genesis to integrate physics-generated data in training Pearl, and innovate on the training methodology in low data regimes. This two-pronged synergistic approach – combining synthetic data and improved training sample-efficiency – represents a significant leap for the future of drug discovery," said Aleksandra Faust, Ph.D., Chief AI Officer of Genesis. "Our evidence of scaling with synthetic data is compelling, and the potential of almost unlimited physics-generated training data is an immense advantage over models constrained to public or costly experimental data."

Genesis rigorously evaluated Pearl against AlphaFold 3 (using reported predictions) and open-source cofolding models that attempt to reproduce AlphaFold 3, including Boltz-1, Boltz-2, Chai-1, and Protenix, using standardized evaluation protocols for accuracy and physical validity of predicted structures:

• Pearl shows the highest performance across all tested models on all benchmarks
• On external benchmarks, Pearl shows up to 40% higher relative performance than AlphaFold 3
• Similarly, Pearl has a substantial advantage on benchmarks from actual Genesis internal drug programs

Pearl was built to supercharge real drug hunters in their everyday workflows. While Pearl already exceeds other models in the pure cofolding setting, Pearl additionally is designed to incorporate expert conditioning during inference. Drug discovery scientists can therefore exploit target-specific data that further improves performance in deployment settings, including on challenging and flexible protein targets.

"AlphaFold 3 was a historic, Nobel-worthy breakthrough, and Pearl is the first model to surpass it in performance," said Evan Feinberg, Ph.D., founder and CEO of Genesis. "Critically, while cofolding models can perform well on certain metrics, it is now well understood in the biopharma industry that they fall short in key real-world settings. These models often fail to truly generalize and sometimes produce obvious physical errors. Our team has focused on developing our foundation model, Pearl, as an integral part of our comprehensive GEMS platform, which uniquely enables drugging difficult targets across different stages in our internal and partner drug programs. This represents a fundamental shift in what’s possible, enabling us to design a new era of medicines that were previously unreachable."

In November 2024, Genesis announced an additional investment from NVentures (NVIDIA’s venture capital arm) and a collaboration to optimize computational methods for AI-powered drug discovery. As part of this collaboration, Genesis has integrated NVIDIA cuEquivariance kernels for triangle operations to accelerate Pearl, achieving 15% relative speedup for training and 10-80% speedup for inference. The companies are also collaborating to further optimize inference operations to enable Pearl’s deployment at even larger scales for Genesis drug programs and partnerships.

"Next-generation foundation models like Pearl, which combine the power of physics and AI, are opening new frontiers in understanding molecular interactions," said Anthony Costa, Director of Digital Biology at NVIDIA. "NVIDIA's accelerated computing platform, featuring libraries like cuEquivariance, is essential for scaling these innovations."

For more information, please read our blog post and the full paper at https://genesis.ml/pearl_technical_report/.

About Genesis Molecular AI

Genesis Molecular AI, Inc. (formerly Genesis Therapeutics) is pioneering foundation models for molecular AI to unlock a new era of drug design and development. The company’s generative and predictive AI platform, GEMS (Genesis Exploration of Molecular Space), integrates AI and physics into industry-leading models to generate and optimize drug molecules, including the breakthrough generative diffusion model Pearl for structure prediction. Genesis has raised over $300 million from leading AI, tech and life science-focused investors, signed multiple AI-focused research collaborations with major pharma partners, and is deploying GEMS to advance an internal therapeutics pipeline for a variety of high-impact targets.

Genesis is headquartered in the San Francisco Bay Area, with locations in San Diego and New York. Learn more at genesis.ml or follow us on LinkedIn.

View source version on businesswire.com:https://www.businesswire.com/news/home/20251028030745/en/

CONTACT: Investors:

Will McCarthy

Chief Operating Officer

[email protected]:

Thermal for Genesis

[email protected]

KEYWORD: UNITED STATES NORTH AMERICA CALIFORNIA

INDUSTRY KEYWORD: RESEARCH TECHNOLOGY HEALTH TECHNOLOGY SOFTWARE BIOTECHNOLOGY PHARMACEUTICAL HEALTH SCIENCE ARTIFICIAL INTELLIGENCE OTHER SCIENCE

SOURCE: Genesis Molecular AI, Inc.

Copyright Business Wire 2025.

PUB: 10/28/2025 02:30 PM/DISC: 10/28/2025 02:30 PM

http://www.businesswire.com/news/home/20251028030745/en