Follow us on Google News (click on ☆)
This is the feat accomplished by chemists from the "Pasteur" laboratory in two publications in PNAS and JACS. By proposing a systematic method to train artificial neurons capable of overcoming the current limitations of molecular simulations, they shed new light on two crucial reactions in prebiotic chemistry.
Chemical reactions are fleeting events in the life of molecules, and observing them directly is a real experimental challenge, even with a powerful "microscope." Molecular simulations are the best tool to follow these events at will. But they encounter two difficulties.
First, following reactive events in real-time means solving Schrödinger's equation* millions of times, which is impossible. Second, even after such immense computational effort, only a handful of observations would be obtained. Yet, to get an accurate picture of a chemical reaction and its thermodynamics, statistical physics teaches us that hundreds of observations are needed because atomic systems are probabilistic.
In two separate studies with similar methodologies, chemists from the PASTEUR laboratory (CNRS/ENS Paris/Paris Sciences et Lettres/Sorbonne University) used an emerging artificial intelligence approach to overcome these limitations. The idea is to train a neural network to solve Schrödinger's equation for all possible structures encountered during a given chemical reaction, thus overcoming the current limitations of molecular simulations.
A critical element to the success of this approach is correctly identifying, selecting, and organizing the relevant data for training. The scientists propose a standardized and robust method tailored for studying chemical reactivity. The result? An approach that bridges the gap between the worlds of quantum physics and statistical physics at a modest computational cost (and therefore a low carbon footprint).
This work, published in PNAS and JACS, paves the way for much more affordable modeling of chemical reactions in condensed phases. The scientists illustrate their methods with two fundamental reactions for life: the formation of a peptide bond necessary for protein synthesis, and the formation of a phosphoester bond for nucleic acid synthesis. They were thus able to highlight the most favorable reaction pathways for these two reactions, which experimental studies cannot directly identify.
These results represent a first step toward understanding the optimal physico-chemical conditions for these highly unfavorable and thus highly improbable reactions. Yet, they led to the formation of the first building blocks of life over four billion years ago, in a context where biological catalysts (enzymes) did not yet exist, and in catalytic conditions that remain to be determined.
*Schrödinger's equation (Erwin Schrödinger, 1925) is a fundamental equation in physics that describes the quantum state of a system.
Author: AVR
References:
Prebiotic chemical reactivity in solution with quantum accuracy and microsecond sampling using neural network potentials.
Z. Benayad, R. David, G. Stirnemann.
Proceedings of the National Academy of Sciences 2024
https://doi.org/10.1073/pnas.2322040121
Competing reaction mechanisms of peptide bond formation in water revealed by deep potential molecular dynamics and path sampling.
R. David, I. Tuñón, D. Laage.
Journal of the American Chemical Society 2024
https://doi.org/10.1021/jacs.4c03445