EvoGen: Generative Protein Design

EvoGen models sequence-to-function relationships across large protein families. It predicts how changes in sequence impact signaling, activity, stability, and immunogenicity. The system designs proteins that meet multiple therapeutic requirements at once, avoiding the tradeoffs common in traditional engineering.

Our work with NVIDIA, published in PNAS, demonstrated steep gains in function and stability using the ProT-VAE architecture, establishing a validated basis for therapeutic design.

EvoLab: High-Throughput Validation

EvoLab synthesizes and assays thousands of proteins in weeks. This includes activity, specificity, biophysical properties, and immunogenicity testing. Experimental data flows directly back into EvoGen, sharpening the model cycle over time.

The microPublication Biology article demonstrates how ML-guided design can outperform natural evolution, reinforcing the platform’s predictive strength.

Deep Data Moat

Every design cycle generates proprietary datasets that refine EvoGen. More data leads to better models, better models lead to higher-quality proteins, and higher-quality proteins open new therapeutic space.

Defensible Advantages

• Proprietary, large-scale performance datasets
• Deep patent estate and IP-protection across the platform and programs
• High-throughput experimental infrastructure
• Validated AI models that improve with each iteration
• Strong collaborations, including NVIDIA and the Gates Foundation.