Traditional antibody discovery has long been a costly, low-efficiency process akin to finding a needle in a haystack. Now, the MoleculeMind MMDesign platform is fundamentally changing this paradigm by achieving a 90.9% target success rate in AI-powered nanobody design. Founded by AI protein folding pioneer Professor Xu Jinbo, the company has successfully shifted biologics design from massive library screening to a highly precise "generate-and-filter" strategy.
MoleculeMind systematically evaluated its platform across 12 high-value therapeutic targets, including cytokines, immune checkpoints, receptor proteins, and multi-transmembrane proteins. By testing only 14 to 50 AI-designed candidate molecules per target via wet-lab expression, the team confirmed specific binding in 11 out of the 12 targets. This extremely low experimental throughput proves the platform's broad applicability across diverse and complex target types.
The platform consistently delivered highly active nanobodies with affinity reaching nanomolar (nM) and picomolar (pM) levels. During tests on the PD-L1 target, the candidate molecule hit rate reached 86.7%. The optimal de novo designed molecule achieved an impressive affinity (KD) of 7.2 nM, while also demonstrating exceptional novelty in its binding conformation space, which establishes deep patent protection for future developments.
Conquering TNF Alpha and GPCR Targets
Beyond conventional targets, the platform achieved landmark results on two of the industry's most challenging categories: TNF alpha and GPCR. TNF alpha presents a shallow, highly solvent-exposed binding interface that previous teams struggled to address with low-throughput AI design. Out of 14 candidates tested, the platform generated 7 specific binding molecules, with the highest affinity reaching an extraordinary 51 pM.
For the GPCR target (CCR7), 22 out of 29 de novo designed nanobodies achieved specific binding. These molecules boasted 90-99% purity and transient expression levels exceeding 0.5 g/L. This level of precision on notoriously difficult targets highlights the maturity of the underlying generative models.
The Era of Programmable Molecular Engineering
The success of the MoleculeMind MMDesign platform signals a definitive end to the "random screening" era in biotech. By achieving a 90.9% success rate with minimal experimental samples, the financial and temporal bottlenecks of animal immunization and large-scale library screening are effectively bypassed. Moving forward, this programmable approach to molecular engineering will likely reduce drug discovery timelines from several years to mere months.
As AI models continue to master complex targets like GPCRs, the barrier to entry for novel biologics will drop significantly. Pharmaceutical companies that fail to integrate generative protein design will quickly find themselves outpaced in both cost-efficiency and patent acquisition. The ability to generate highly specific, patentable molecules with just a few dozen wet-lab tests is no longer a theoretical goal - it is the new industry baseline.
