This AI Designs Drugs in Minutes

Why It Matters

The breakthrough could dramatically accelerate early‑stage drug discovery, reducing costs and making previously intractable targets druggable.

Key Takeaways

• Ligendal’s Ligan Forge creates 700 peptide sequences per second.
• Model runs 10,000× faster than Boltzgen, million× faster than Bcraft.
• Generated 150,000 candidates for five hard targets in 3.4 minutes.
• Traditional tools missed all targets; Ligan Forge succeeded on all.
• Discrete diffusion model learns physics of molecular interactions directly.

Summary

On March 17, 2026, Andre Watson, a biomeaterials scientist and founder of Ligendal, released a preprint describing a new AI system that designs peptide drugs in minutes.

The system, called Ligan Forge, uses a discrete diffusion model that learns the physics of molecular interactions and generates peptide sequences directly from the shape of a target protein’s docking site. It can produce over 700 sequences per second on a single GPU—roughly 10,000 times faster than the industry‑standard Boltzgen and a million times faster than Bcraft.

In benchmark tests on five of the most challenging drug targets, Ligan Forge generated 150,000 candidate peptides in just 3.4 minutes, while Boltzgen succeeded on only one target and the competing tool Minecraft produced none. The results demonstrate that conventional docking tools struggle with these hard targets, whereas the diffusion model consistently finds viable binders.

If the approach scales, it could compress years of laboratory work into hours, slashing R&D costs and opening the door to rapid iteration on previously “undruggable” proteins. Investors and pharma companies are likely to watch Ligendal closely as the technology promises to reshape early‑stage drug discovery.

Original Description

It produces over 700 sequences per second on a single GPU, and tested on 5 hard drug targets, LigandForge generated 150,000 candidates in 3.4 minutes, succeeding where other tools failed.