Accelerated Discovery of BPA-Free Dental Resin Candidates through Active Learning of Network-Emergent Properties
- 1Department of Material Design and Innovation
- 2University at Buffalo
- 3Department of Restorative Dentistry
Abstract
Bisphenol A-glycidyl methacrylate (Bis-GMA) is widely used in dental resins, but its BPA-derived chemistry motivates the search for structurally distinct monomers that retain favorable cured-network properties. Here we introduce EloKriti, an active-learning workflow that links molecular candidate representations to explicit in silico curing and multi-objective optimization of network-emergent properties. We showcase EloKriti by navigating a curated 1000-candidate BPA-free chemical space, filtered from Open Macromolecular Genome for compatibility with vinyl chain-growth polymerization, to simultaneously maximize elastic modulus and minimize volumetric shrinkage and fractional free volume. The retrospectively normalized dominated hypervolume increased from 0.371 to 0.590 by iteration 6, when 57 candidates had been successfully evaluated; although additional candidates were evaluated in the next two rounds, the hypervolume and Pareto-set size did not increase further. From the resulting 13-member Pareto front, three chemically distinct backbones were subsequently incorporated into established, industry-relevant resin building blocks through candidate-specific reactive-site mapping consistent with the growth chemistry of each building block. The resulting synthetically actionable monomers were subjected to five-replica molecular-dynamics virtual-curing validation, and all three networks successfully reached the 95% target degree of conversion. These virtually cured networks exhibited highly specialized property profiles: one optimized the density–shrinkage–free-volume balance, another maximized elastic modulus, and the third achieved the highest glass-transition temperature. These results provide synthetically actionable BPA-free leads for future experimental evaluation and a data-efficient route for prioritizing chemistries on the basis of properties that emerge only after network formation.
Related
- Harnessing Machine Learning for Accelerated Drug Discovery: Opportunities and Unmet Challenges
- Computationally Accelerated Discovery of Efficient Dual Component White Light Emitters
- Robot-Accelerated Perovskite Investigation and Discovery: Copper Halide Perovskitoids
- Crystalline ASR products: Review of mineral candidates and molecular simulations of their atomic structures, thermodynamic data and thermoelastic properties
- Graph Neural Networks for Accelerated Prediction of Excited-State Properties in Transition-Metal Complexes
BibTeX
@misc{balantrapu2026accelerated,
title = {{Accelerated Discovery of BPA-Free Dental Resin Candidates through Active Learning of Network-Emergent Properties}},
author = {Sai Harshit Balantrapu and Sabrina Feitosa Sochacki and Ganesh Sivaraman},
howpublished = {ChemRxiv},
year = {2026},
month = sep,
doi = {10.26434/chemrxiv.15008229/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15008229/v1},
note = {CC BY-NC-ND},
}
RIS
TY - UNPB
TI - Accelerated Discovery of BPA-Free Dental Resin Candidates through Active Learning of Network-Emergent Properties
AU - Sai Harshit Balantrapu
AU - Sabrina Feitosa Sochacki
AU - Ganesh Sivaraman
PY - 2026
DA - 2026/09/02
PB - ChemRxiv
DO - 10.26434/chemrxiv.15008229/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15008229/v1
AB - Bisphenol A-glycidyl methacrylate (Bis-GMA) is widely used in dental resins, but its BPA-derived chemistry motivates the search for structurally distinct monomers that retain favorable cured-network properties. Here we introduce EloKriti, an active-learning workflow that links molecular candidate representations to explicit in silico curing and multi-objective optimization of network-emergent properties. We showcase EloKriti by navigating a curated 1000-candidate BPA-free chemical space, filtered from Open Macromolecular Genome for compatibility with vinyl chain-growth polymerization, to simultaneously maximize elastic modulus and minimize volumetric shrinkage and fractional free volume. The retrospectively normalized dominated hypervolume increased from 0.371 to 0.590 by iteration 6, when 57 candidates had been successfully evaluated; although additional candidates were evaluated in the next two rounds, the hypervolume and Pareto-set size did not increase further. From the resulting 13-member Pareto front, three chemically distinct backbones were subsequently incorporated into established, industry-relevant resin building blocks through candidate-specific reactive-site mapping consistent with the growth chemistry of each building block. The resulting synthetically actionable monomers were subjected to five-replica molecular-dynamics virtual-curing validation, and all three networks successfully reached the 95% target degree of conversion. These virtually cured networks exhibited highly specialized property profiles: one optimized the density–shrinkage–free-volume balance, another maximized elastic modulus, and the third achieved the highest glass-transition temperature. These results provide synthetically actionable BPA-free leads for future experimental evaluation and a data-efficient route for prioritizing chemistries on the basis of properties that emerge only after network formation.
ER -