Gero
GeroHacking aging
with physics and AI
Platform · Product 01

ProtoBind-Diff

A diffusion model for de novo ligand design. Generate target-conditioned, binding-competent molecules directly from a protein sequence, without needing a solved co-crystal structure.

Request access → See Harvest benchmark
How it works

From sequence to candidate molecule.

ProtoBind-Diff conditions a diffusion process on a target's protein sequence and samples novel molecules that are predicted to bind. Because it works from sequence rather than structure, it reaches targets where no experimental structure exists yet.

The output is a diverse set of candidate ligands ready for downstream scoring, filtering, and synthesis planning.

ProtoBind-Diff workflow

Capabilities

What the model is built to do well.

01
Sequence-conditioned
Generate against a target from its protein sequence alone, opening up structurally uncharacterized targets.
02
De novo diversity
Sample chemically diverse, novel scaffolds rather than retrieving near-neighbors of known binders.
03
Pipeline ready
Outputs flow straight into scoring and filtering, including evaluation with the Harvest benchmark.
Validation
Validation

Measured, not asserted.

ProtoBind-Diff is evaluated against open benchmarks so its performance can be checked independently. We publish the methodology and release the work for the community to reproduce.

See how we benchmark →
Resources

Get the details.

Scientific backgroundThe physics-and-AI thesis behind the model Harvest benchmarkHow we measure discovery models Request accessinfo@gero.ai