Predicting compositional changes of organic–inorganic hybrid materials with Augmented CycleGAN

Qianxiang Ai,Joshua Schrier,Alexander Norquist

doi:10.1039/d1dd00044f

Qianxiang Ai, Joshua Schrier + Show 1 more

Open Access

https://doi.org/10.1039/d1dd00044f

Copy DOI

Abstract

Image-to-image translation models applied to materials: augmented CycleGAN models for predicting chemical compositions of hybrid materials.

Highlights

Organic–inorganic hybrid crystalline materials are a wide class of functional materials that encompasses halide perovskites,[1–3] metal organic frameworks (MOFS),[4,5] and templated metal oxides.[6]
We describe the generation of amine-templated metal oxides (ATMOs) compositions through Augmented CycleGAN,[38] a novel generative model that can learn many-to-many relations between two domains through unpaired data
We focus on an analogous composition translation problem for aminetemplated metal oxides (ATMO, see Methods for detailed de nitions): given the chemical compositions of structures containing amine A, can we learn a function that transforms them to compositions of structures containing amine B? As a speci c example, we chose amine A and amine B to be N-methylmethanamine (SMILES: CNC) and ethane-1,2-diamine (SMILES: NCCN), respectively, as they are the two amines found most frequently in the Cambridge Structural Database (CSD) as of 2021

Summary

Introduction

Organic–inorganic hybrid crystalline materials are a wide class of functional materials that encompasses halide perovskites,[1–3] metal organic frameworks (MOFS),[4,5] and templated metal oxides.[6]. The great structural diversity found in ATMOs (exempli ed by the amine-templated zinc phosphate structures of four different dimensionalities), can only be matched by their compositional diversity (71 elements, 25 main group building units, and 349 amines as of 2021).[12]. This immense chemical space, along with various types of possible interactions, makes it extremely challenging to predict the properties of novel ATMOs. Since the seminal works on generative adversarial networks[13] (GAN) and variational autoencoder[14] (VAE) in 2014, generative models have proliferated in multiple disciplines, including biology,[15] geology,[16] and meteorology.[17].

Objectives

Methods

Findings

Conclusion