Research and Development: User projects

Flexible and intrinsically disordered proteins (IDPs) are crucial for signal transduction in biological systems. Their hallmark is their diverse structural conformations. While experimental methods yield averaged attributes of IDPs, they fall short in depicting the dynamic spectrum of these conformations. Molecular dynamics (MD) simulations offer an alternative, providing detailed snapshots of IDPs at atomic or residue resolution. Due to IDPs' complex ensemble nature, comprehensive simulations are necessary to traverse the multiple energy troughs they inhabit, making their simulation more resource-intensive than that of structured proteins. Shaw and colleagues have recently shown the efficacy of all-atom MD simulations in mapping the varied structures of IDPs, albeit at significant computational expense. An efficient substitute, like the physics-based United-RESidue (UNRES) force field, can significantly reduce computational demands while still exploring IDP conformational diversity effectively as demonstrated in our recently published work. In UNRES, ⍺-carbons dictate the backbone geometry of amino acid chains, with aggregated peptide groups and side chains forming the interaction points. Its energy function represents a mean force potential in aqueous conditions, simplifying secondary structural elements. Our work has validated the UNRES model's predictive prowess for structured proteins, as evidenced by its commendable performance in the CASP trials. When refined with experimental data, UNRES could potentially enhance the precision of generating structure ensembles of IDPs.

Nuclear Magnetic Resonance (NMR) chemical shifts, due to their high precision and wide condition applicability, have become a gold standard for the structural analysis of flexible proteins. Integrating these experimental measures with computational models enhances prediction accuracy and helps bridge knowledge gaps. Our newly created database and the associated software, Glutton, can swiftly navigate through the conformational space, outpacing all-atom approaches by a factor of a thousand. Glutton can discern multiple structural preferences from a single set of chemical shifts, indicating residues that toggle between states. We are updating the Glutton database, and its capacity has tripled over the past five years. Guided by machine-learning techniques, Glutton can predict backbone torsional angles directly from chemical shift data. UNRES has already included an API to include torsional angle restraints in the current state. Our strategy involves enhancing the conformational search by incorporating torsional restraints from Glutton into the UNRES framework. We aim to utilize multiplexed replica exchange molecular dynamics (MREMD) for its proven efficiency in sampling conformational space. An additional energy term will be introduced to improve the representation of disordered proteins in our simulations. We plan to conduct a detailed cluster analysis to identify and define a representative structural ensemble, which will consist of approximately 10,000 unique configurations.

Our objective is to fine-tune the UNRES model to proficiently and precisely map the conformational landscape of IDPs. We plan to refine UNRES simulations guided by dihedral angle distributions derived from NMR chemical shift information via Glutton. To achieve these ends, we seek access to the Tryton supercomputer and its storage facilities. This resource will enable us to calibrate and evaluate the UNRES model's performance in real-life scenarios.

The TASK supercomputer hosts the most advanced version of UNRES, which is crucial for our project's success. The computational capabilities of the Tryton supercomputer are indispensable for the fine-tuning and extensive assessment of the refined UNRES force fields for practical applications.

The research described above is supported by grant Sheng-2 awarded to the PI (grant contract number: UMO-2021/40/Q/ST4/00035) and by grants of the U.S. collaborator.