The GenSci148 program illustrates a shift from one-time mutation prediction to iterative protein engineering under real drug-development constraints
SHANGHAI, Aug. 20, 2026 /PRNewswire/ — Shanghai Matwings Technology Co., Ltd. (“Matwings”) today announced that GenSci148 Injection, an investigational ophthalmic biologic developed by Changchun GeneScience Pharmaceutical Co., Ltd. (“GenSci”), has received clinical trial clearance in China for neovascular age-related macular degeneration (nAMD), diabetic macular edema (DME) and retinal vein occlusion (RVO).
Matwings supported the GenSci148 program through four iterative AI–wet lab cycles covering 222 protein variants. Rather than applying AI as a one-time mutation-prediction tool, the teams repeatedly used experimental results to redesign both molecules and optimization objectives. The program progressed from improving biological activity to balancing potency, stability, expression, formulation and other development-relevant properties—illustrating a shift from predicting mutations to engineering therapeutic proteins under real drug-development constraints.
Four AI–Wet Lab Cycles, 222 Variants
Matwings and GenSci implemented a closed-loop workflow combining AI-guided molecular design, wet-lab testing, experimental feedback and redesign.
Across four iterative cycles, the teams evaluated 222 protein variants. The first two rounds assessed 92 and 30 variants, respectively, with an initial focus on biological activity. The next two rounds evaluated 50 variants each, expanding the optimization objectives to include high-concentration formulation viscosity and broader developability requirements.
Experimental results from each round were incorporated into subsequent design cycles, allowing not only the molecular designs but also the optimization objectives to evolve as evidence accumulated.
The campaign therefore progressed from:
activity optimization → broader developability optimization → multi-objective molecular engineering
Across the optimization campaign, experimentally tested variants demonstrated improvements or favorable performance across six development-relevant properties:
Development Parameter
Observed Result
VEGF-A binding affinity
Up to approximately 10-fold improvement
VEGF-A/C/D functional
blockade
Up to approximately 3-fold improvement
Nonclinical in vivo activity
Inhibitory activity observed 84 days after dosing in the
evaluated retinal model
Thermal stability
Tm increased by up to approximately 4.5°C
Protein expression
Increased by up to approximately 27.6%
High-concentration formulation
viscosity
Reduced by approximately 13 cP
Together, these results reflect optimization across three increasingly demanding dimensions of therapeutic protein development: biological performance, molecular developability and formulation-relevant properties.
In the nonclinical retinal model evaluated during the program, the optimized molecule maintained inhibitory activity 84 days after dosing. Under the specific experimental conditions tested, it also showed greater inhibition than aflibercept and faricimab.
These findings are preclinical and do not establish comparative clinical efficacy or safety.
Beyond Mutation Prediction: Engineering Proteins Under Real Drug-Development Constraints
Therapeutic protein engineering requires balancing biological activity with stability, expression, formulation and other development constraints. In the GenSci148 program, experimental results continuously informed subsequent AI-guided designs, enabling the optimization process to move beyond a single fixed objective.
Some experimentally validated mutations were located away from the conventional target-binding interface, illustrating how AI-guided exploration can identify productive regions of sequence space beyond interface-focused design.
The program also illustrates a broader progression in protein AI validation:
computational benchmarking → experimental validation → repeated integration within a real drug-development workflow
GenSci148 represents the third stage of this progression, with AI-guided design and experimental evidence repeatedly linked across successive engineering cycles.
The Venus Protein AI Stack
Matwings’ Venus protein AI stack has evolved from sequence-centered modeling toward systems integrating three-dimensional structure, evolutionary information and task-specific capabilities. Importantly, the AI-guided engineering work supporting GenSci148 was conducted using Venus 1.0, while Venus has since advanced to Venus 3.0, represented by VenusREM, which integrates protein sequence, three-dimensional structure and evolutionary information for mutation-effect prediction.
Rather than relying on a single model, different components are applied to different protein R&D and engineering tasks within the broader closed-loop workflow.
Generation
Representative Model(s)
Information Integrated
Primary Role
Venus 1.0
Venus 1.0
Protein sequence
Sequence-
centered
modeling
Venus 2.0
Venus-ProSST
Sequence + 3D structure
Structure-aware
modeling
Venus 3.0
VenusREM
Sequence + 3D structure +
evolutionary information
Mutation-effect
prediction
Task-Specific
Models
Venus-FSFP, Venus-Maxwell, Venus-Mine, Venus-RXN, Venus-
Fold
Task-dependent
Protein R&D and
engineering
Certain models within the broader Venus portfolio are in development or planned stages.
Together, these models form a specialist AI stack that can be combined with experimental data and iterative design across different stages of protein engineering.
From External Validation to Therapeutic Creation
The GenSci148 collaboration provides external validation of Matwings’ protein-engineering capabilities in a real therapeutic development program.
In March 2026, Matwings established Shanghai Biowings Therapeutics Co., Ltd. (“Biowings Therapeutics”) to apply the same closed-loop engineering approach to internally originated therapeutic programs. Matwings develops the underlying AI and protein-engineering technology engine, while Biowings Therapeutics combines these capabilities with disease biology and drug-development expertise to create and advance therapeutic candidates.
“The real test of AI for science is not whether a model performs well on a benchmark, but whether it can create measurable value through repeated design–experiment cycles in an actual R&D program,” said Prof. Liang Hong, Founder and Chief Scientist of Matwings. “GenSci148 marks an important step from demonstrating individual model capabilities toward building a repeatable protein-engineering system.”
“Matwings contributed important molecular engineering and optimization capabilities to the GenSci148 program,” said Dr. Lei Jin, CEO of GenSci. “Rather than relying on one-time predictions, the teams repeatedly combined AI-guided design with experimental evidence to improve properties relevant to drug development.”
About Matwings
Shanghai Matwings Technology Co., Ltd. is an AI-driven protein R&D company integrating specialist AI models, computational molecular design and experimental validation to support iterative protein engineering and drug development.
For more information, visit www.matwings.com.
About Biowings Therapeutics
Shanghai Biowings Therapeutics Co., Ltd. focuses on AI-enabled therapeutic asset creation and clinical translation.
Scientific and Development Notice: GenSci148 is an investigational product. Its safety and efficacy have not been established. Preclinical findings are specific to the experimental systems and conditions evaluated and may not predict clinical outcomes.
Contact:
Website: www.matwings.com
E-mail: public@biowingsthera.com
View original content:https://www.prnewswire.com/news-releases/matwings-demonstrates-closed-loop-protein-ai-through-four-aiwet-lab-cycles-and-222-protein-variants-302856288.html
SOURCE Shanghai Matwings Technology Co., Ltd.; Shanghai Biowings Therapeutics Co., Ltd.
Technology4 days ago
Coin Market4 days ago
Coin Market4 days ago
Technology3 days ago
Technology3 days ago
Technology4 days ago
Technology2 days ago
Technology4 days ago