Matwings Reports Four-Cycle Closed-Loop Protein AI Program

Shanghai Matwings Technology announced on August 20, 2026, that its GenSci148 program completed four iterative AI–wet-lab cycles involving 222 protein variants.

The workflow used experimental results from each cycle to redesign both protein sequences and optimization objectives, moving beyond one-time mutation prediction.

The first two cycles evaluated 92 and 30 variants, focusing mainly on biological activity; the next two tested 50 variants each and added formulation viscosity and broader developability criteria.

Reported improvements included up to about 10-fold better VEGF-A binding affinity, 3-fold stronger VEGF-A/C/D functional blockade, a thermal-stability increase of up to 4.5°C, expression gains of up to 27.6%, and a viscosity reduction of about 13 cP.

In an evaluated retinal model, the optimized molecule reportedly maintained inhibitory activity 84 days after dosing and showed greater inhibition than aflibercept and faricimab under the stated test conditions.

The announcement is a company press release, so the reported performance should be viewed as preliminary until independently validated and published in peer-reviewed or regulatory documentation.