Matwings Advances Protein Engineering Through Iterative AI Cycles

The collaboration moved beyond simple mutation prediction, treating artificial intelligence as a continuous component of the drug discovery workflow. By feeding experimental data back into the design process, the teams evolved their optimization objectives from basic biological activity to include stability, expression levels, and formulation viscosity. Over the course of four cycles, this iterative approach resulted in a 10-fold improvement in VEGF-A binding affinity and a 27.6% increase in protein expression. In retinal models, the optimized molecule demonstrated sustained inhibitory activity for 84 days.

Scaling the Venus AI Stack

This development utilized the Venus 1.0 platform, a precursor to the company's current Venus 3.0 system. While early iterations relied primarily on sequence-centered modeling, the technology has expanded to incorporate three-dimensional structural data and evolutionary information. Matwings recently launched a subsidiary, Biowings Therapeutics, to apply these closed-loop engineering methods to its internal drug pipeline. According to Prof. Liang Hong, founder and chief scientist at Matwings, the success of GenSci148 demonstrates that the true value of AI in science lies in its capacity for repeatable engineering cycles rather than isolated performance benchmarks.