2020/01/16 by Aditi S. Krishnapriyan, Maciej Harańczyk, Krishnapriyan, Aditi S. +3 · 3 citations
Biochemistry, Genetics and Molecular Biology · Computer Science · #Algebraic Topology (math.AT) #Bioinformatics and Genomic Networks #Computational Physics (physics.comp-ph) #FOS: Computer and information sciences #FOS: Mathematics #FOS: Physical sciences #Geochemistry and Geologic Mapping #Machine Learning (cs.LG) #Materials Science (cond-mat.mtrl-sci) #Topological and Geometric Data Analysis
paper · pdf · doi:10.48550/arxiv.2001.05972
openalex publication_date 2020/01/16 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Machine learning has emerged as an attractive alternative to experiments and\nsimulations for predicting material properties. Usually, such an approach\nrelies on specific domain knowledge for feature design: each learning target\nrequires careful selection of features that an expert recognizes as important\nfor the specific task. The major drawback of this approach is that computation\nof only a few structural features has been implemented so far, and it is\ndifficult to tell a priori which features are important for a particular\napplication. The latter problem has been empirically observed for predictors of\nguest uptake in nanoporous materials: local and global porosity features become\ndominant descriptors at low and high pressures, respectively. We investigate a\nfeature representation of materials using tools from topological data analysis.\nSpecifically, we use persistent homology to describe the geometry of nanoporous\nmaterials at various scales. We combine our topological descriptor with\ntraditional structural features and investigate the relative importance of each\nto the prediction tasks. We demonstrate an application of this feature\nrepresentation by predicting methane adsorption in zeolites, for pressures in\nthe range of 1-200 bar. Our results not only show a considerable improvement\ncompared to the baseline, but they also highlight that topological features\ncapture information complementary to the structural features: this is\nespecially important for the adsorption at low pressure, a task particularly\ndifficult for the traditional features. Furthermore, by investigation of the\nimportance of individual topological features in the adsorption model, we are\nable to pinpoint the location of the pores that correlate best to adsorption at\ndifferent pressure, contributing to our atom-level understanding of\nstructure-property relationships.\n