Intelligent Molecular Identification Approach to High-Efficiency Solvents for Organosulfide Capture Using the Active Machine Learning Framework

There is increasing interest in the development of intelligent
strategies for rational design and/or identification of promising
solvent compounds for volatile, environmentally unfriendly compound
capture. However, typical computer-aided methods require huge datasets
and/or suffer from accumulated prediction bias. Here, we constructed
a computational framework by introducing a stepwise screen approach
for molecular descriptors and molecular active selection machine learning
to modify the adequate chemical space iteratively. This framework
identifies the optimal solvent candidates by molecular similarity
search and iterative molecular addition to the training dataset. In
a virtual screening of 126,068 compounds, 2443 solvent candidates
were successfully identified for the capture of methyl mercaptan (MeSH),
one of the major organosulfides in fossil gases.