Two MIT-trained researchers have built a web platform that lets laboratory scientists run protein engineering models directly from a browser, eliminating the need for specialized hardware and programming skills. Pharmaceutical firm Boehringer Ingelheim adopted the system in early 2025 for oncology, immunology, and inflammatory disease programs, marking a concrete move of this academic tool into commercial drug development.
A browser-based solution for lab biologists
Co-founders Tristan Bepler and Tim Lu designed OpenProtein.AI to fix a common friction point in biological research. Bepler, who earned a PhD in computational biology from MIT in 2020, later joined Lu's lab as a postdoctoral researcher. Lu noticed that biologists frequently needed advanced machine learning but rarely had the software engineering background to implement it.
"We wanted to build something that was user friendly because machine-learning ideas are kind of esoteric," Bepler said. Standard machine learning setups demand GPUs, careful hyperparameter tuning, and custom sequence libraries-infrastructure many labs lack. The platform abstracts these details. Scientists can upload experimental data, train predictive models, generate protein variants, and validate results without leaving a standard browser.
How PoET reduces data requirements
The platform's engine is PoET (Protein Evolutionary Transformer), a model trained on global evolutionary datasets to recognize patterns in amino acid mutations across related species. By screening digital protein variants, researchers can skip thousands of physical wet-lab iterations.
Key operational specs of the model:
- 30x less experimental data needed to start training, so models become useful with minimal initial lab results.
- 182 million parameters-compact enough for high-efficiency processing without heavy computational overhead.
- Generates and screens variant sequences to narrow candidates before any pipette testing.
A 2025 update, PoET-2, showed that focused, domain-specific models can equal or beat much larger AI systems while using a fraction of the training data and compute power. The platform's efficiency makes it a practical tool in settings where big-team resources are not available, aligning with the broader category of AI for Science & Research.
Enterprise use and the Boehringer Ingelheim deal
OpenProtein.AI offers its platform at no cost to university labs and charges commercial licensing fees to biotech and pharma companies. Boehringer Ingelheim signed on in early 2025, embedding the generative models into its therapeutic protein engineering work across several disease areas. The partnership has since expanded, with the pharma team using the system directly within its research pipeline.
Parallel institutional efforts are applying similar transformer architectures at a larger scale. A collaboration between the Arc Institute, NVIDIA, Stanford, UC Berkeley, and UCSF produced Evo2, a genomic foundation model trained on 9.3 trillion nucleotides from 128,000 genomes. Published in Nature in 2026, Evo2 targets disease-causing mutations and synthetic genome design across all domains of life. While PoET tackles rapid protein engineering for individual labs, macro-scale models like Evo2 reflect a shift toward treating DNA and protein structures as programmable languages.
Why this matters for Science and Research professionals
For wet-lab biologists and biochemists, the practical effect is a lower bar to using machine learning. A platform that works inside a browser and needs 30x less experimental data lets scientists train models on their own results without waiting for a computational specialist. The Boehringer Ingelheim adoption shows that pharma teams are betting on these tools to shorten the gap between early molecular hypotheses and validated drug candidates, meaning researchers who become fluent with protein design models now will be positioned to move faster as drug development timelines compress.
Your membership also unlocks:
