The world of biology is on the cusp of a revolutionary transformation, and at the heart of this change is the fusion of artificial intelligence (AI) with protein design. This article delves into the fascinating work of OpenProtein.AI, a company that is empowering biologists with cutting-edge AI tools, and explores the implications and potential of this fusion.
Unlocking the Power of AI for Biologists
AI has already demonstrated its ability to accelerate drug development and enhance our understanding of diseases. However, translating AI into tangible treatments requires making the latest AI models accessible to scientists. The challenge lies in the fact that most scientists are not machine-learning experts.
OpenProtein.AI steps in to bridge this gap with a no-code platform. This platform provides scientists with access to powerful foundation models and a suite of tools for protein design, structure prediction, and model training. The company's mission is to ensure that scientists, regardless of their background, can stay at the forefront of AI-driven biology.
Advancing Biology with AI: A Personal Journey
Tristan Bepler, co-founder of OpenProtein.AI, embarked on a journey at MIT in 2014, driven by a realization: our understanding of biomolecules and proteins was not sufficient for accurate predictive models. This led him to explore ways to predict amino acid chains using evolutionary data, even before AlphaFold, Google's powerful protein structure prediction model, was released.
Bepler's work resulted in one of the first generative AI models for protein design, a protein language model. He was fascinated by the classical framework of proteins and the complex relationships between their sequence, structure, and function. His goal was to use foundation models to bypass the 'structure' component and go directly from sequence to function.
Democratizing AI for Biologists
After joining Tim Lu's lab at MIT as a postdoc, Bepler found himself at the intersection of AI and biology. He recognized the potential of AI to accelerate scientific work but also the gap between cutting-edge tools and biologists who lacked coding expertise.
OpenProtein.AI was born out of the idea to broaden access to these powerful tools. The platform features an intuitive web interface, allowing biologists to upload data and conduct protein engineering with machine learning. It offers a range of open-source models, including PoET, OpenProtein's flagship protein language model.
The Power of PoET and Beyond
PoET, or Protein Evolutionary Transformer, is trained on protein groups to generate related proteins. It can generalize evolutionary constraints and incorporate new protein sequence information without retraining, enabling researchers to improve the model with their experimental data.
The model helps researchers design and analyze proteins, generating libraries of sequences in silico and using predictive models for validation. It's a no-code front-end, but OpenProtein also provides APIs for those who prefer coding.
Enabling the Future of Therapies
OpenProtein's platform has gained traction, with collaborations like the one with Boehringer Ingelheim, a large pharmaceutical company. The company's latest model, PoET-2, outperforms larger models while using fewer resources and data.
The founders are now focused on describing proteins in a meaningful, domain-specific language, incorporating evolutionary constraints, and understanding enzymatic reactions.
The Future of AI-Driven Biology
As AI progresses, OpenProtein aims to create models that consider the dynamic and interconnected nature of protein function. The company's mission is to ensure that AI resources remain accessible, preventing concentration and ensuring progress in the scientific field.
In my opinion, the work of OpenProtein.AI is a testament to the power of AI in biology and its potential to revolutionize drug development and our understanding of diseases. It's an exciting time, and I believe we are witnessing the birth of a new era in biological research.
