Alladi Ramakrishnan Hall

A Radical Geometric Alternative Complementary to MonteCarlo(MC) and Molecular Dynamics(MD) for Pair-Potential driven soft-matter Assembly Modeling

Meera Sitharam

University of Florida

Multiscale modeling, both data-driven (e.g. machine learning) and mechanistic (e.g.molecular dynamics MD or monte carlo MC) are now routine for equilibrium modeling of (macro)molecular or particle cluster assembly in diverse scientific scenarios. They take as input: (i) geometry of the assembling units (flexibly tethered rigid collections of atom centers), and (ii) the assembly-driving pair-potentials (internal) energy (or corresponding force field), incorporating an implicit solvent. The outputs are (a) binding affinity and hotspot residue prediction, (b) paths and kinetic transition network generation and (c) optimal assembly landscape design solutions. Overall, while prevailing methods have not reached the limits of accuracy for meeting the modeling challenges, they have reached the limits of efficiency and scalability of algorithms and data representations, storage and movement, to take advantage of the availability of massive amounts of data and computational resources.

The talk will propose to address these challenges by building on the recent, distinctive EASAL methodology (Efficient Atlasing, Search and sampling of Assembly Landscapes). While complementing prevailing approaches, the methodology exploits landscape geometry/topology, through the lens of recent and ongoing developments in discrete geometry and rigorous algorithmic complexity analysis. EASAL is a scale-independent algorithm suite with a stand-alone, opensource prototype implementation - under vigorous ongoing development. Current work building on EASAL has demonstrated the methodology's markedly distinct capabilities:(i) minimal, refinable representation of potential-energy basin topology, permitting fast path sampling and kinetic transition network generation (ii) upto 3000X coverage efficiency against prevailing MC sampling for helix assembly and against a competing free energy method for particle cluster assembly; and
(iii) multiscaling via hybridization of mechanistic and data-driven methods, leading to a rapid, data-driven EASAL prediction of hotspot residues for virus capsid assembly that was experimentally validated.

[ssl.gstatic.com/ui/v1/icons/mail/images/cleardot.gif]