Machine learning (ML) models can be interpretable, providing new fundamental physical/chemical insights into materials systems. We developed interpretable ML models¹ to investigate two very different materials systems:

• Peptide “wires” experimentally shown to exhibit long-range charge transport. Thermal fluctuations have been implicated as important in this emergent property.¹

• We combined large-scale DFT calculations of 10³ peptide dimer snapshots together with ML feature analysis to determine peptide regions most relevant for conductivity-associated electronic structure features.³
• We used ML-guided feature analysis to determine the most important dimer molecular conformation parameters for controlling charge transport.³

• Mg-Y alloys with deformation twinning, of interest as lightweight structural materials⁴

• Combined large-scale microstructural characterization studies of multiple Mg-Y alloys together with ML classification and feature analysis to elucidate the processing parameters that are statistically significant for predicting the presence of a deformation twin within a grain.
• Discovered that only four processing parameters are necessary for prediction of twinning with > 80% accuracy.

¹ Mastracco, et al. ACS Nano 2022.
² Lewis, et al. JPC C 2022. 
³ Mastracco, Mohanam, Nagaro et al. J. Chem. Inf. Model. 2025
⁴ Mastracco, Yu et al.In revision, Adv. Eng. Mater.2025