For a modest 100-amino-acid protein, there are 10^130 possible sequences, while all life on Earth has only explored ~10^43. This vast, unexplored space means we can now design binders for "undruggable" targets that evolution never needed to create.
Related Insights
The relationship between a multi-specific antibody's design and its function is often non-intuitive. LabGenius's ML platform excels by exploring this complex "fitness landscape" without human bias, identifying high-performing molecules that a rational designer would deem too unconventional or "crazy."
AlphaFold's success in identifying a key protein for human fertilization (out of 2,000 possibilities) showcases AI's power. It acts as a hypothesis generator, dramatically reducing the search space for expensive and time-consuming real-world experiments.
A-muto suggests many drug programs fail due to toxicity from hitting the wrong epitope, not a flawed biological concept. By identifying and targeting a structural epitope unique to the diseased state of the same protein, these previously abandoned but promising therapies could be salvaged.
Frances Arnold, an engineer by training, reframed biological evolution as a powerful optimization algorithm. Instead of a purely biological concept, she saw it as a process for iterative design that could be harnessed in the lab to build new enzymes far more effectively than traditional methods.
Current AI for protein engineering relies on small public datasets like the PDB (~10,000 structures), causing models to "hallucinate" or default to known examples. This data bottleneck, orders of magnitude smaller than data used for LLMs, hinders the development of novel therapeutics.
Instead of screening billions of nature's existing proteins (a search problem), AI-powered de novo design creates entirely new proteins for specific functions from scratch. This moves the paradigm from hoping to find a match to intentionally engineering the desired molecule.
The AI-discovered antibiotic Halicin showed no evolved resistance in E. coli after 30 days. This is likely because it hits multiple protein targets simultaneously, a complex property that AI is well-suited to identify and which makes it exponentially harder for bacteria to develop resistance.
ProPhet's strategy is to focus on 'hard-to-drug' proteins, which are often avoided because they lack the structural data required for traditional discovery. Because ProPhet's AI model needs very little protein information to predict interactions, this data scarcity becomes a competitive advantage.
Beyond optimizing existing biological functions, Frances Arnold's lab uses directed evolution to create enzymes for entirely new chemical reactions, like forming carbon-silicon bonds. This demonstrates that life's chemical toolkit is a small subset of what's possible, opening up a vast "non-natural" chemical universe.
Generative AI alone designs proteins that look correct on paper but often fail in the lab. DenovAI adds a physics layer to simulate molecular dynamics—the "jiggling and wiggling"—which weeds out false positives by modeling how proteins actually interact in the real world.