Periodic Reporting for period 1 - bAIes (Integrative, AI-aided Inference of Protein Structure and Dynamics)
Periodo di rendicontazione: 2023-08-01 al 2026-01-31
We are living in an exciting moment for the life sciences. Artificial Intelligence (AI) tools, most notably AlphaFold2, have revolutionized our ability to predict protein structures from amino acid sequence with remarkable speed and accuracy. Yet knowing the three-dimensional structure of a protein is only the beginning. Proteins are not static objects: they move, change shape, and respond to their cellular environment. This dynamic behavior is central to how proteins carry out their functions in health and disease, but they cannot be captured by structure-prediction tools alone.
This project addresses that gap. We will develop bAIes, a new modelling approach that combines the strengths of AI, advanced molecular simulations, and experimental data. By doing so, we aim to describe not just what proteins look like, but also the full range of conformations they can adopt, how they interconvert between these states, and how their environment influences these processes. This will open the door to tackling problems that current AI methods cannot address, such as understanding flexible and disordered protein regions or studying proteins directly within their natural cellular context.
The bAIes approach will be made freely available through PLUMED, a widely used open-source software platform, ensuring that researchers worldwide can apply it to a broad range of biological challenges. The expected outcome is a versatile and accessible tool that pushes the boundaries of what is possible with AI in biology, helping the scientific community to better understand the molecular machinery of life.
This project addresses that gap. We will develop bAIes, a new modelling approach that combines the strengths of AI, advanced molecular simulations, and experimental data. By doing so, we aim to describe not just what proteins look like, but also the full range of conformations they can adopt, how they interconvert between these states, and how their environment influences these processes. This will open the door to tackling problems that current AI methods cannot address, such as understanding flexible and disordered protein regions or studying proteins directly within their natural cellular context.
The bAIes approach will be made freely available through PLUMED, a widely used open-source software platform, ensuring that researchers worldwide can apply it to a broad range of biological challenges. The expected outcome is a versatile and accessible tool that pushes the boundaries of what is possible with AI in biology, helping the scientific community to better understand the molecular machinery of life.
This project will move beyond current strategies where AI-predicted protein structures are sequentially refined against individual experimental datasets. Such approaches risk discarding valuable information and propagating errors. Instead, we will develop bAIes, a modelling framework that treats AI models as a new type of experimental data and integrates them probabilistically with experimental measurements and molecular simulations. The main achievement will be the creation of a Bayesian framework that combines all sources of information in a single step, providing accurate models of protein structures and dynamics without privileging one dataset over another. We will apply bAIes to problems that remain out of reach for current AI methods, including the study of flexible and disordered protein regions and the determination of structures and dynamics directly in their cellular environment. The final outcome will be a versatile and accessible computational approach, released through the open-source PLUMED library, enabling the broader community to tackle a wide range of biological problems with unprecedented accuracy.
This project will deliver an integrative modelling approach to uncover the molecular mechanisms that underlie biological functions. By combining AI-predicted structures, experimental measurements, and molecular simulations, bAIes will make it possible to characterize not only static protein structures but also the populations of conformational states and the dynamic pathways connecting them. This represents a significant step beyond the current capabilities of AI-based structure prediction, providing a more complete and realistic picture of how proteins work in their natural context. The impact of bAIes will be twofold. Scientifically, it will allow researchers to address questions that remain out of reach today, from the role of disordered protein regions to the behavior of proteins inside cells. Practically, bAIes will accelerate discoveries across fields such as structural biology, molecular medicine, and drug design by offering a rigorous, efficient, and accessible tool to model protein dynamics. To maximize adoption by the community, bAIes will be distributed through PLUMED, a widely adopted open-source software library, ensuring immediate access by the international research community. All protocols and data will be shared via PLUMED-NEST, fostering transparency and reproducibility. To maximize the impact, further uptake will benefit from continued community-driven development, training initiatives to broaden expertise in integrative modelling, and sustained support for open science practices. Together, these efforts will establish bAIes as a reference framework for the next generation of protein modelling.