Identifying optimal reaction coordinates in complex molecular systems with Genetic Programming
Project ID: 2228cd1281 (You will need this ID for your application)
Research Theme: Engineering
UCL Lead department: Chemical Engineering
Lead Supervisor: Eric Fraga
Project Summary:
In particle-based simulations of complex molecular systems, a significant portion of the computational intricacies lie in identifying slowly changing descriptors of the system’s state, i.e. being able to identify if a given configuration should be classified as a reactant, product, or transition state.
Such descriptors are referred to as collective variables (CVs). These CVs play a crucial role in both understanding and enhancing rare, activated transitions between metastable states of interest. While chemical/physical intuition plays an essential role in identifying meaningful CVs, often, intuition provides sub-optimal candidates that fail to appropriately describe the reaction coordinates associated with complex collective events such as crystal nucleation, self-assembly, and protein binding/unbinding.
In this project, we aim to investigate the use of genetic programming for model identification and characterization, aiming to obtain optimal non-linear combinations of physically intuitive CVs to identify maximally informative, low-dimensional approximations of the natural reaction coordinate. We will explore the application of different criteria for optimality, singly and together in a multi-objective senses, including the ability of candidate “optimal CVs” to approximate the committor probability function, their ability to maximise the spectral gap associated with the system’s dynamics, and their ability to preserve the Markovian character of activated transitions between metastable states.
Genetic programming will initially make use of existing optimization methods, including meta-heuristic nature-inspired methods suitable for the computationally expensive evaluation of the criteria itemised above. The project will also address the extension and adaptation of these optimization methods for effective and efficient search of the space of the non-linear combinations of CVs.
The student will receive training in optimization, molecular dynamics, and programming.