91探花

Drug discovery has long been defined by time, cost and uncertainty. Developing and optimising a single therapeutic antibody can take years of iterative experimentation, with researchers testing countless variations before identifying a viable candidate. Even incremental improvements to existing drugs often require extensive lab work and dedicated R&D campaigns. But a new experiment suggests that artificial intelligence may be starting to compress that process dramatically.

A recent test by biotech startup is offering a glimpse into how that shift might unfold. The company revealed that its antibody design platform, ConvergeAB鈩�, was able to generate an improved version of cetuximab, a widely used cancer therapy, using a single prompt and no task-specific training. Within eight hours, the system produced an antibody sequence that demonstrated more than double the binding strength of the original drug in lab validation tests.

Cetuximab is commonly used to treat colorectal, head, and neck cancers by targeting the epidermal growth factor receptor (EGFR), a protein that drives tumor growth. The effectiveness of the drug depends heavily on how tightly it binds to this receptor. Stronger binding can translate into more effective blocking of cancer-promoting signals.

In this case, Converge Bio鈥檚 AI-generated version showed an average binding affinity approximately 2.1 times stronger than cetuximab and 4.4 times stronger than a competing solution, based on surface plasmon resonance (SPR) testing. The design required only six targeted modifications to the original antibody sequence, changes that were distributed across both structural and functional regions.

What makes the result notable is not just the improvement itself, but how it was achieved. The system was run in a 鈥渮ero-shot鈥� setting, meaning it was not trained specifically for this task. Instead, researchers provided the model with the original antibody and the target receptor sequence, and the system generated roughly 100,000 potential candidates. From those, a small subset was selected for laboratory validation, with only 10 sequences ultimately tested.

The findings suggest that AI can move beyond assisting drug discovery into actively redesigning existing therapies with minimal human intervention. Rather than relying on traditional iterative methods, where researchers test one modification at a time, models like ConvergeAB can explore vast design spaces simultaneously and converge on high-performing candidates far more efficiently.

鈥淭his work began as an internal experiment to see how far our platform could go, and our results reinforce what is possible in therapeutic development through AI,鈥� said Dov Gertz, CEO and co-founder of Converge Bio. 鈥淲hat stands out is both the quality of the results and the pace at which we were able to reach them. In just eight hours, we moved from a single prompt to a set of candidates that demonstrated meaningful improvements, giving a solid example of how AI can compress what has traditionally been a long and intensive process into a far shorter timeframe.鈥�

The company has filed a provisional patent for the AI-designed antibody sequence.

While the results are still early and limited to preclinical validation, they point to a broader shift underway in drug development. Instead of treating AI as a support tool for analysing biological data, companies are beginning to position it as a core engine for molecular design capable of generating and optimising drug candidates before they ever reach the lab.

This approach could significantly reduce the cost and experimental burden associated with drug discovery. In the Converge Bio experiment, only a handful of candidates needed to be physically tested, compared to the much larger experimental campaigns typically required in antibody optimisation.

Converge Bio describes its long-term vision as a 状computational lab,状 where hypotheses are generated, tested, and refined digitally before being validated experimentally. If realised at scale, that model could fundamentally change the economics and timelines of bringing new therapies to market.