An intelligent agent for general-purpose protein engineering

Proteins offer an exciting path to address a multitude of biotechnological challenges.

Capable of working under non-toxic, mild conditions and performing a myriad of functions, their controllable design has been sought-after for decades.

However, to gain a technological advantage in a world with pressing demands in sustainability and healthcare, we must accelerate the development of custom-tailored, proficient proteins.

In this proposal, we will develop an intelligent system capable of efficiently engineering functional proteins tailored to user-defined specifications.Artificial Intelligence (AI) advancements are promoting a fresh wave of enthusiasm across many fields, providing solutions to problems that escape human intuition.

Recently, protein language models (pLMs) are showing unprecedented performance in generating novel, efficient proteins. We have trained three advanced pLMs, demonstrating promising preliminary results in experimental settings.

In this proposal, we will train an agent that will learn from combined sequence, structural, functional, and dynamic data to perform multiple protein engineering tasks.

The agent will iteratively improve from experimental feedback using Reinforcement Learning, and explainable AI will allow us to ‘open the black box’ and understand its decision process.

A vital component of this work will be its rigorous experimental validation, progressing through increasingly challenging tasks with biotechnological applications.

This project will deliver an intelligent agent with continuous learning capabilities, accessible through user-friendly interfaces, empowering researchers worldwide with an easy-to-use tool to design custom-tailored proteins.

In addition, by incorporating explainability, it will offer a novel angle to understanding complex sequence-to-function relationships.

Lastly, comprehensive experimental validation will assess the reliability and applicability of these novel approaches in real-world contexts.