Integrative, AI-aided Inference of Protein Structure and Dynamics

Introduction

The life sciences community is living in exciting times. During the past year, Artificial Intelligence (AI), and in particular AlphaFold2, has contributed to advancing our understanding of protein behaviour by enabling structure prediction with accuracy comparable to many experimental techniques at a fraction of their time and costs.

Need for Comprehensive Understanding

However, structures are only a piece of the puzzle. To understand the mechanisms underlying biological functions, we need to:

1. Characterize the conformational landscape of proteins.
2. Identify the population of relevant states.
3. Understand their pathways of interconversion.

Furthermore, we need to determine the effect of the environment in modulating structures, populations, and pathways, as biological systems perform their functions in the complexity of cells rather than in the isolation of test tubes. None of these objectives can be achieved by AI structure-prediction methods alone.

Proposed Solution

In this proposal, we will leverage the PI’s expertise in the field of integrative computational-experimental techniques to develop, apply, and disseminate bAIes, a modelling approach that will enable attaining these goals.

Methodology

bAIes will make synergistic use of:

• AI structural models
• Experimental data
• Molecular simulations driven by accurate physico-chemical models

This combination will allow us to characterize protein structure and dynamics effectively.

Applications and Focus

We will demonstrate how bAIes can solve biological problems that exceed the capabilities of AI approaches, such as:

• The characterization of protein disordered regions
• The determination of structure and dynamics in situ, with a particular focus on the SARS-CoV-2 spike protein.

Expected Outcomes

The outcome of this proposal will be a versatile, accurate, and efficient method that will push the boundaries of what can be achieved with AI structure-prediction methods.

Implementation

bAIes will be implemented in the widely used PLUMED library, of which the PI is founder and core developer, thus enabling its application to a wide variety of systems and biological problems beyond those envisioned here.