MAMABIO: A Scientific Breakthrough in Biomass Transformation

A project at the heart of the bioeconomy

Part of the B-BEST program, the MAMABIO project aims to remove barriers to the use of data from atomic-scale modeling for designing processes that transform bio-based molecules. Specifically, it focuses on leveraging machine learning methods to model and predict, at the atomic scale, the speed and selectivity of reactions transforming bio-based molecules.

Building on developments initiated over several years at the University of Lorraine and Comenius University in Bratislava, the project successfully tackled one of the major challenges in this field: the prohibitive cost of atomic-scale quantum calculations (known as ab initio calculations) required to accurately predict reaction rates. For the first time, it was possible to combine the methodological challenges of ab initio molecular dynamics (AIMD) with the complexities of using a high level of theoretical rigor.

What are ab initio calculations and ab initio molecular dynamics (AIMD)?

Ab initio calculations predict the collective behavior of atoms by solving the Schrödinger equation. This can be done at various levels of theoretical rigor: the higher the level of theory, the more precise the result, but also the greater the computational cost. AIMD is an advanced technique that additionally accounts for the effects of temperature on reaction rates. These techniques are used to predict reaction rate constants in catalytic systems, such as zeolites (materials that accelerate chemical reactions).

To achieve satisfactory precision in predicting reaction rates, it is necessary to combine high-level theoretical calculations with AIMD, which incurs a prohibitive computational cost. For instance, a single reaction rate constant calculation at the so-called RPA (Random Phase Approximation) theoretical level within AIMD could take about a millennium!

*The RPA approach resolves the Schrödinger equation at a high theoretical level and estimates the electronic correlation energy—the energy of interaction between electrons due to their mutual influence.

A concrete advancement for fuel and chemical molecule production

Faced with this challenge, the MAMABIO team adopted an innovative approach by integrating Machine Learning Perturbation Theory (MLPT), a method applied for the first time this year to a catalytic reaction at such a high theoretical level. Their work, published in Angewandte Chemie, International Edition (featured in Science) and expanded upon in a second publication in Catalysis, Science and Technology, demonstrated the potential of this method for predicting reaction rate constants. This opens new horizons for modeling chemical reactions in various fields, particularly for transforming bio-based molecules in energy and chemical applications.

The MAMABIO team is now using AIMD to study the transformation of specific alcohols into molecules useful for fuel and chemical production. Specifically, the dehydration of alcohols into alkenes (a type of hydrocarbon)—a key reaction in the production of fuels and chemicals from biomass—is being investigated. AIMD has proven critical for understanding the transformation of isobutanol into linear alkenes, as shown by recently published results in ACS Catalysis in collaboration with Comenius University. The MAMABIO team now plans to apply the MLPT method to these types of reactions.

More accurate predictions for optimized processes

Thanks to this new methodology, the team achieved unprecedented accuracy in modeling chemical reactions, surpassing so-called "static" approaches. This advancement not only enables more accurate reaction predictions but also allows for tailoring catalysts to the specific needs of biomass transformation processes. Based on this foundation, the MAMABIO project now aims to predict the dehydration rates of alcohols into various possible products across multiple catalyst types to construct predictive kinetic models of performance. Experimental studies are also being conducted within the project to validate these predictions and provide complementary data for higher-scale model construction.

The MAMABIO project thus opens new avenues for developing more efficient and sustainable processes, contributing to the transition toward cleaner energy and the valorization of bio-based molecules. This scientific breakthrough could have significant impacts across various industrial sectors, particularly those related to the bioeconomy and sustainable fuels.

Publication references for further reading:

Jérôme Rey, Céline Chizallet, Dario Rocca, Tomáš Bučko, Michael Badawi. Reference-Quality Free Energy Barriers in Catalysis from Machine Learning Thermodynamic Perturbation Theory. Angewandte Chemie, International Edition, 2024, 63, e202312392. ⟨10.1002/anie.202312392⟩.⟨hal-04504311⟩

Jérôme Rey, Michael Badawi, Dario Rocca, Céline Chizallet, Tomáš Bučko, Machine learning thermodynamic perturbation theory offers accurate activation free energies at the RPA level for alkene isomerization in zeolites. Catalysis Science and Technology, 2024, 14, 5314-5323. ⟨10.1039/D4CY00548A⟩.

Monika Gešvandtnerová, Pascal Raybaud, Céline Chizallet, Tomáš Bučko. Importance of Dynamic Effects in Isobutanol to Linear Butenes Conversion Catalyzed by Acid Zeolites Assessed by AIMD. ACS Catalysis, 2024, 14 (10), pp.7478-7491. ⟨10.1021/acscatal.4c00736⟩. ⟨hal-04592530⟩

IFPEN, actualité : « L’apprentissage machine accélère l’accès par dynamique moléculaire ab initio à des données de haute précision pour la chimie », 25/07/2024.