Gero Launches ProtoBind-Diff, a Structure-Free AI Foundation Model for Targeted Small Molecule Discovery

• ProtoBind-Diff generates drug-like molecules for specific protein targets using only their amino acid sequences—no 3D structures required.
• The model performs competitively with leading structure-based tools in predicting binding strength, while also generating novel and chemically diverse compounds.

SINGAPORE, June 25, 2025 (GLOBE NEWSWIRE) -- Gero, a biotechnology company focused on aging and chronic diseases, today announced the launch of ProtoBind-Diff, a masked diffusion language model that generates small molecules conditioned solely on protein sequences. Trained on more than one million active protein–ligand pairs, ProtoBind-Diff represents a paradigm shift in molecular generation. Unlike structure-based models, which are limited by the small and biased set of resolved protein–ligand complexes, ProtoBind-Diff leverages the vastly larger body of activity data available in public databases. This enables training on a much broader chemical and biological space, helping the model generalize to underexplored targets where structural data is sparse or unavailable.

Gero released a new preprint on bioRxiv detailing the model’s performance and design.

“Designing small molecules that hit protein targets is one of the hardest problems in drug discovery. Classical modeling struggles because the energy scales, polarization effects, and the complexity of protein dynamics make high-resolution predictions nearly impossible. But maybe we’ve been asking the wrong question,” said Peter Fedichev, Ph.D., CEO and Co-Founder, Gero. “Nature had to solve this puzzle already - evolution optimized a biochemical language that encodes how proteins and molecules interact. With ProtoBind-Diff, we’re tapping into that. It’s a language model that learns from sequences, not structures. It doesn’t simulate physics-it learns the grammar of bioactivity from a million real examples.”

ProtoBind-Diff was developed as a foundational component of Gero’s generative drug discovery platform. The model leverages pre-trained protein embeddings (ESM-2) and a denoising diffusion framework to generate chemically valid and novel molecules in SMILES format, guided by sequence-level information alone.

Key results from the preprint include:

• Competitive performance with structure-based models (e.g., Pocket2Mol, TargetDiff) in structure-aware benchmarks using Boltz-1, a neural network that predicts protein–ligand complexes and scores their binding quality. ProtoBind-Diff matches or exceeds these models in both well-characterized (“easy”) and low-data (“hard”) targets.
• Emergent interpretability, with attention heads aligning to known binding residues despite no exposure to 3D binding site annotations during training.
• High novelty, drug-likeness, and synthesizability of generated molecules, as measured by structural similarity, drug-likeness, and synthesizability metrics.
• Open-source release available on GitHub, with a waiting list for the public demo of the full model and codebase at https://github.com/gero-science/ProtoBind-Diff.

ProtoBind-Diff was benchmarked using both classical docking methods (AutoDock Vina) and structure-aware deep learning models. ProtoBind-Diff was benchmarked using Boltz-1, an open source neural network inspired by AlphaFold 3, the development of the Nobel Prize–recognized breakthrough in protein structure prediction. Boltz-1 extends this capability to model how proteins bind small molecules, offering a scalable, structure-aware metric for evaluating binding strength. The model consistently demonstrated strong enrichment for active compounds, particularly in targets with minimal structural data or few known ligands. In some cases, its Boltz-1 enrichment factors exceeded those of structure-trained models, suggesting a robust ability to learn spatial priors from sequence embeddings, highlighting the ability to learn spatial priors from sequence alone.

“I believe we are only at the beginning of the journey toward creating an ideal generative model. Yes, in our benchmarks, the ProtoBind-Diff model outperforms some existing 3D structural models,” said Konstantin Avchaciov, Ph.D., Senior Researcher at Gero and lead scientist behind the project. “That said, I am confident that as we continue to expand our datasets to include a broader diversity of protein classes, we will achieve significantly better results in the future.”

The release of ProtoBind-Diff aligns with growing interest in human-relevant, structure-agnostic approaches to drug discovery, particularly in areas like pandemic response, neglected disease targets, and proteins with intrinsically disordered regions.

Gero has integrated ProtoBind-Diff into its internal discovery pipeline and is actively seeking partners to apply the model in collaborative programs across oncology, immunology, infectious disease, and aging-related conditions.

About Gero
Gero is a biotechnology company advancing novel therapeutics for age-related diseases and longevity. The company combines proprietary biological datasets with AI-driven models to understand and slow the aging process and ultimately extend healthy human lifespan. Gero is also collaborating with Pfizer to develop treatments for fibrotic diseases as part of its broader mission to target the root causes of aging. To learn more, visit gero.ai.

Media Contact
Kimberly Ha
KKH Advisors
917-291-5744
kimberly.ha@kkhadvisors.com

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.