Machine-Learning-Guided Discovery and Optimization of Additives in Preparing Cu Catalysts for CO₂ Reduction

Abstract

Discovery and optimization of new catalysts can be potentially accelerated by efficient data analysis using machine-learning (ML). In this paper, we record the process of searching for additives in the electrochemical deposition of Cu catalysts for CO2 reduction (CO2RR) using ML, which includes three iterative cycles: “experimental test; ML analysis; prediction and redesign”. Cu catalysts are known for CO2RR to obtain a range of products including C1 (CO, HCOOH, CH4, CH3OH) and C2+ (C2H4, C2H6, C2H5OH, C3H7OH). Subtle changes in morphology and surface structure of the catalysts caused by additives in catalyst preparation can lead to dramatic shifts in CO2RR selectivity. After several ML cycles, we obtained catalysts selective for CO, HCOOH, and C2+ products. This catalyst discovery process highlights the potential of ML to accelerate material development by efficiently extracting information from a limited number of experimental data.

Feasibility and application of machine learning enabled fast screening of poly-beta-amino-esters for cartilage therapies

“…Majority of research dedicated to implement ML in drug discovery/chemistry employs a very narrow range of potential models, even just one [49][50][51] , without a clear rationale for the selection of the algorithms included in the pool assessed 5,18,[52][53][54][55] . Here instead, we purposely screened a large number of potential algorithms based on different approaches (e.g.…”

Recent progress and prospect of electrodeposition-type catalysts in carbon dioxide reduction utilizations

“…Recently, Wang et al have provided a new path for additive selection and optimization by combining experiment and theory with efficient data analysis through machine guidance (Figure 6c) [43] . They indicate that Sn salt can be used as an important additive for the production of CO and There are many influencing factors of electrodeposition, due to its huge technical parameter space.…”

Interpretable Machine Learning of Two-Photon Absorption

“…564 of the features come from molecular fragment fingerprint (MFF) featurization. 34 In MFF, molecular fragments were generated by the extended-connectivity fingerprints (ECFP) method using a radius of 436 supported by the Deepchem python toolkit. 35 A vector recording the appearance times of each fragment in a molecule 36 was then created (Figure 2).…”