Back to search

Georgia Institute of Technology

On Efficient Screening Strategies for Molecular Structures

Abstract

dc:description.abstract

The accelerating demand for functional materials across structural, environmental, and polymeric applications has rendered traditional experimental and high-fidelity simulation workflows computationally intractable at discovery-relevant scales, motivating the development of data-driven screening strategies that are simultaneously efficient, physically grounded, and generalizable to data-scarce regimes. This work introduces a suite of physics-informed machine learning frameworks for molecular-scale structure-property (S-P) linkage and autonomous candidate screening, addressing three classes of molecular complexity of increasing difficulty: perfect and disordered crystalline alloys, small molecules including per- and polyfluoroalkyl substances (PFAS), and polymeric systems. We first employ the Molecular Knowledge System (MKS) framework to encode DFT-computed electron charge density (ECD) fields as voxelized two-point spatial autocorrelations, and demonstrate that these reduced-order structural descriptors, when coupled with lean convolutional neural networks (CNNs) of only ~80,000 fittable parameters, accurately predict bulk modulus and total energy across chemically diverse crystal datasets while training in under 20 minutes on a single GPU. We then extend this MKS-based representation into an active learning loop for extrapolative alloy discovery in the TiNbAlZr quaternary composition space, where uncertainty-guided Gaussian process regression (GPR) surrogates accurately recover target properties using a minimal number of actively selected DFT samples from a candidate space of over 18,000 structures, bypassing more than 99.85% of prospective DFT calculations, with demonstrated generalization to 5-, 6-, and 7-component alloy systems through iterative model refinement. Shifting to small-molecule complexity, we develop molecular screening workflows for PFAS compounds in which spatially-aware voxelized molecular structure (VMS) features, encoding intrinsic elemental attributes including electronegativity and ionization energy, are coupled with GPR to predict thermodynamically relevant global properties such as exact polarizability and reaction free energies, while graph-topology features fed into random forest models enable accurate prediction of C-F bond dissociation energies (BDEs), directly quantifying the kinetic barriers governing PFAS degradation. Finally, to bridge the fidelity gap between atomistic accuracy and the computational efficiency of coarse-grained (CG) simulation, we formulate a Bayesian optimization (BO) protocol that autonomously calibrates CG Martini3 force-field bonded topology parameters against all-atom molecular dynamics (AAMD) targets, specifically density and radius of gyration, converging to AAMD-fidelity for polyethylene and polystyrene homopolymers, and successfully reproducing macroscale phase-separation morphology in PS/PMMA and PVDF/PMMA polymer blends. Taken together, this thesis establishes that low-dimensional, physics-grounded structural quantification via spatial statistics and graph-based features, combined with computationally lean surrogate models and intelligent design-of-experiments strategies, enables reliable, extrapolative, and scale-bridging molecular screening while remaining fully compatible with the data-scarce and resource-constrained conditions ubiquitous in practical materials discovery pipelines.

Degree

thesis:*
Name thesis:degree_name
Mechanical Engineering, PhD
Department thesis:degree_department
Mechanical Engineering
Grantor
Georgia Institute of Technology
Year dc:date.issued
2026

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Ray, Pranoy
Advisor dc:contributor.advisor
  • Kalidindi, Surya R.
Committee members dc:contributor.committeemember
  • Kumar, Satish
  • Deo, Chaitanya
  • Medford, Andrew
  • Kolel-Veetil, Manoj
ORCID person:identifier.orcid

Rights

Language dc:language.iso
English

Identifiers

dc:identifier.*
Handle dc:identifier.uri
https://hdl.handle.net/1853/82149

Chain of custody

source
Harvested from
Georgia Tech
Base URL
repository.gatech.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Ray, Pranoy. On Efficient Screening Strategies for Molecular Structures. Georgia Institute of Technology, 2026. https://hdl.handle.net/1853/82149