Machine learning of spectra-property relationship for imperfect and small chemistry data

Abstract

Machine learning (ML) is causing profound changes to chemical research through its powerful statistical and mathematical methodological capabilities. However, the nature of chemistry experiments often sets very high hurdles to collect high-quality data that are deficiency free, contradicting the need of ML to learn from big data. Even worse, the black-box nature of most ML methods requires more abundant data to ensure good transferability. Herein, we combine physics-based spectral descriptors with a symbolic regression method to establish interpretable spectra–property relationship. Using the machine-learned mathematical formulas, we have predicted the adsorption energy and charge transfer of the CO-adsorbed Cu-based MOF systems from their infrared and Raman spectra. The explicit prediction models are robust, allowing them to be transferrable to small and low-quality dataset containing partial errors. Surprisingly, they can be used to identify and clean error data, which are common data scenarios in real experiments. Such robust learning protocol will significantly enhance the applicability of machine-learned spectroscopy for chemical science.

Spectroscopy-Guided Deep Learning Predicts Solid–Liquid Surface Adsorbate Properties in Unseen Solvents

“…Some previous studies have indicated that the performance of a trained model in a single-solvent system including gas or water phase degrades significantly once it is directly transferred to unseen systems or scenarios. , Therefore, to make sure that HMNN is omniscient to various solvents, we design a subsequent learning process (tier II) to extract solvent difference knowledge on a few-sample and multisolvent data set . During tier II, for each individual task, we select the top two important experts that have already learned more than 90% of the required task knowledge in total from both gas and water phases and freeze the four selected experts (2 from gas phase and 2 from water phase) for subsequent network assembling (Figure A).…”

“…9 Additionally, using a regression-based method facilitates learning an approximate formulaic mapping of QSPR, hence addressing the obstacle of imperfect and limited data. 10 However, regarding the learned end-to-end relationships, their nature of low dimension and lack of chemical explanatory factors determine the inevitably limited generalization abilities of DL-based models.…”

Catalytic Structure Design by AI Generating with Spectroscopic Descriptors

“…physically meaningful quantity, the spectra have been used directly as descriptors for predicting catalytic properties. 22,23 However, the implementation of dynamic structural inversion and continuous catalytic generation based on spectroscopic descriptors is still to be developed.…”

“…For mining the implied spectral information, machine learning (ML) has given a glimpse of achieving breakthroughs. − ML can be used to exact useful features from spectra, to establish correlations between spectra and catalyst properties or structures and to optimize catalyst design based on spectral feedback. As a physically meaningful quantity, the spectra have been used directly as descriptors for predicting catalytic properties. , However, the implementation of dynamic structural inversion and continuous catalytic generation based on spectroscopic descriptors is still to be developed.…”

Structure-Based Multilevel Descriptors for High-throughput Screening of Elastomers

“…ML techniques have significantly advanced the establishment of high-throughput screening (HTS) pipelines for discovering a broad spectrum of materials. − This progress is evident in the polymer domain, primarily attributed to the evolution of polymer informatics, which has introduced a diverse set of descriptors for numerically characterizing polymers, including fingerprint descriptors, physiochemical descriptors, and graph descriptors. − Given the complex nature of polymers, in terms of their structure and composition, continuous efforts are being made to enhance these descriptors to better represent and predict polymer properties. For the purpose of predicting a range of polymer properties, Ramprasad et al developed a hierarchical fingerprint encompassing three distinct scales: atomic, quantitative structure–property relationship, and morphological levels .…”