MoleculeMind Claims 94.7% Project Success Rate in AI Protein Design

Plenty of teams can now generate thousands of protein sequences. MoleculeMind (分子之心 in Chinese) proposes measuring something else: how many molecules have to go to experimental validation, and how often a project ends with a working candidate. In a new technical report on its AI design platform MMDesign, the company cites a 94.7% project success rate across 18 targets. All six GPCR targets in the set yielded a hit.

Why attrition, not generation, became the key metric

In AI-driven biologics development, the bottleneck shifted long ago. A model can produce as many variants as you like, but every experimental cycle costs money and weeks of work: synthesis, expression, affinity measurement. If only a handful of a thousand submitted molecules turn out to work, the project's economics fall apart, no matter how impressive the generation looks.

That is why MoleculeMind shifts the emphasis to the "entry into the lab": the fewer candidates need to be tested to succeed, the cheaper each project is. The 94.7% figure, counted per project rather than per molecule, reflects exactly this logic. The platform is judged as a pipeline, not as a generator.

Six GPCRs as a stress test

The most eye-catching part of the report concerns GPCRs. These G protein-coupled receptors account for a significant share of approved drugs. For biologics, however, they remain a difficult target. The receptor sits in the membrane, its extracellular regions are small and mobile, and the purified protein is often unstable. Antibodies against such targets turn out inconsistently.

That makes a set of six GPCR targets, with the platform getting a hit on every one, more convincing than success on "convenient" soluble proteins. It is a move from isolated demonstrations to systematic validation, as the report's title puts it.

What exactly was tested

Key parameters of the report:

• platform: MMDesign, an AI system for de novo molecular design;
• developer: MoleculeMind (分子之心);
• scope: 18 targets;
• modalities: nanobodies (VHH) and macrocycles;
• share of successful projects: 94.7%;
• GPCR targets: all six yielded a hit;
• report publication date: October 27, 2026.

The combination of two modalities is also telling. VHHs are single-domain antibodies. They are more compact than classical immunoglobulins and reach hard-to-access regions of targets well. Macrocycles sit between small molecules and biologics. A single platform that handles both protein and cyclic structures is less dependent on any one format.

How to read claims like this

The report was published by the company itself, and there is no independent verification yet. So the market will want details that determine what 94.7% is worth:

• how a "project success" is defined: binding at the level of a few candidates, functional activity, or affinity in the desired range;
• how many molecules had to be synthesized and tested per target;
• how novel the targets are relative to the training data and known antibodies;
• whether there are results beyond "hits": selectivity, stability, suitability for further development.

For pharma companies and investors, this is a practical checklist. If the answers to these questions prove strong, MMDesign will become an argument that AI design of biologics is moving from demonstrations to a predictable process. For teams building similar platforms themselves, the benchmark is now clear: compare the share of successful projects and the number of molecules sent to experiments, not the volume of generation.