Beyond AlphaFold: AI Is Starting to Design New Biology

AlphaFold transformed biology by predicting the three-dimensional structures of proteins. A new system called AlphaProtein Novo takes a very different step: instead of predicting proteins that already exist, it attempts to design entirely new enzymes with functions requested by researchers.

The system uses machine learning to build both protein structure and amino-acid sequence around a desired catalytic reaction. In its first demonstrations, it created enzymes capable of performing new-to-nature chemical reactions, including highly selective synthesis of piperidine structures widely used in drug development.
AlphaProtein Novo also designed enzymes capable of degrading DEHP, an environmental chemical, under conditions that disabled naturally occurring enzymes. In several benchmark reactions, the designed enzymes achieved catalytic performance comparable with or better than established alternatives, and the researchers report that de novo design could outperform simply searching nature for suitable enzymes in some difficult problems.

The immediate work is not about cancer, but the implications for cancer research could be substantial. Artificially designed enzymes might eventually be created to activate drugs selectively inside tumours, destroy metabolites cancer cells depend on, modify the tumour microenvironment, or perform chemical reactions that biology does not normally possess.
The important shift is conceptual. AI is moving from predicting biological structures and screening existing molecules towards designing new biological functions from scratch. AlphaProtein Novo is still preclinical research but it provides another indication of how rapidly AI-driven drug and protein design is moving from observation toward engineering.

Source.