Machine Learning in Biomolecular Simulations

MACHINE LEARNING IN BIOMOLECULAR SIMULATIONS

LEAD PI (S): Prof. Dr. Paolo Carloni & Prof. MICHELE PARRINELLO

MENTOR: DR. ANDREA RIZZI

WEBSITE: https://www.fz-juelich.de/en/ias/ias-5/research/advances-in-molecular-dynamics/machine-learning-accelerated-molecular-simulations

This project develops techniques at the interface between machine learning and physics-based molecular simulations to improve the accuracy and efficiency of alchemical free energy calculations of protein-ligand systems of pharmacological relevance. It builds on a method designed to overcome the problem of slow convergence between the end states of the alchemical transformation^1,2  .

Depending on their interests and skills, the student will work and acquire expertise on (a subset of) the following topics:

• Alchemical binding free energy calculations.
• Normalizing flows neural networks.
• Targeted free energy estimation.
• High-performance computing on highly parallel architectures.
• Enhanced sampling techniques (e.g., REST2, metadynamics).
• Hybrid quantum mechanics/molecular mechanics (QM/MM) calculations.
• Development and/or application of hybrid machine learning/molecular mechanics (ML/MM) potentials.

We anticipate that the work will lead to a publication in a high-impact journal, and opportunities for future follow-ups and longer-term collaborations between the groups will be abundant.

COMPUTATIONAL RESOURCES

During the internship, the student will have access to the computational resources at Jülich Forschungszentrum³ and RWTH Aachen University IT Center⁴, which include CPU and GPU cluster computing systems among the fastest and most energy-efficient in the world. Forschungszentrum Jülich represents one of the main supercomputing centers in Europe, provides training and assistance on high-performance computing, and will host the first European exascale supercomputer by the end of 2023.

REQUIRED SKILLS

Student applying to this project are expected to have the following skills:

• Basic knowledge in physical chemistry and experience with molecular simulations (classical and/or quantum mechanics).
• Basic knowledge of programming (Python is preferred).

Nice to have:

• Familiarity with the PyTorch library and with basic machine learning and neural network concepts.
• Experience with alchemical free energy calculations.

References and Links

1. Rizzi A, Carloni P, Parrinello M. Targeted free energy perturbation revisited: Accurate free energies from mapped reference potentials. The Journal of Physical Chemistry Letters. 2021;12(39):9449-54.
2. Jarzynski C. Targeted free energy perturbation. Physical Review E. 2002;65(4):046122.
3. https://www.fz-juelich.de/ias/jsc/EN/Expertise/Supercomputers/supercomputers_node.html
4. https://www.itc.rwth-aachen.de/go/id/eucm/?lidx=1

Please also look at our website for updates.