BURNABY, NOVEMBER 18, 2023 — In a scientific first, researchers at Anodyne have used deep learning-powered AI, quantum chemistry, and molecular modeling to design de novo enzymes — never-before-existing proteins that accelerate biochemical reactions.
Enzymes drive a wide range of critical processes, from photosynthesis to digestion. Anodyne, collaborating with Prof. Roberto Chica at the University of Ottawa, used their newly developed computational enzymology platform to create a new kind of transferase enzyme. Transferases — as their name implies — transfer specific functional groups from one molecule to another and are involved in hundreds of different biochemical pathways.
“Enzymes are remarkable chemists,” said Manou Davies, Anodyne’s co-founder and CSO. “Rather than relying on toxic compounds or extreme heat to catalyse reactions, enzymes can break down or build up whatever we need. By creating new enzymes capable of undertaking the chemical transformations we need to replace fossil fuels, we put more renewable chemicals and fuels within reach.”
Deep learning design: There’s been a recent glut of headlines about AI-powered creations — especially text generators, like ChatGPT, or image generators, like DALL-E. But AI has also found a place in discovering new chemistries, especially for large complex molecules like enzymes.
High-powered AI is a key component of enzyme design because it can predict winning design combinations much faster than individuals. Enzymes are extremely complex molecules, honed by millions of years of evolution, so the number of possible intricate and important structures that can make all the difference to what an enzyme does is difficult to comprehend. This makes the standard approach of enzyme design both cost- and labour-prohibitive.
By combining deep learning, quantum chemistry, and molecular modeling, Anodyne was able to perform virtual directed evolution to reduce the astronomical number of possible design alternatives (1035), into a small combinatorial library to experimentally test (~100). In-house mutagenesis and high throughput screening was then used to create and validate an entirely new-to-the-world transferase enzyme.
The bigger deal: The team sees their work as potentially bigger than creating a new enzyme, however. With the AI-powered method proven, the door is now open to creating a wide array of enzymes capable of simplifying chemical pathways, or making products out of reach of traditional fermentation. Artificial enzymes could create improved biofuels, serve as medical diagnostic tools, or break down plastics and pollutants.
“We are now able to design very efficient enzymes from scratch on the computer, as opposed to relying on enzymes found in nature,” Davies said. “This achievement represents a significant step forward in our enzyme development capabilities and means that custom enzymes for almost any chemical reaction could, in principle, be designed to revolutionize the feedstocks for industrial, chemical and personal care products.”