Prediction of Molecular Electronic Transitions Using Random Forests

Abstract

<div>Fluorescent molecules, fluorophores, play essential roles in bioimaging. Attachment</div><div>of fluorophores to proteins enables observation of the detailed structure and dynamics</div><div>of biological reactions occurring in the cell. Effective bioimaging requires fluorophores</div><div>with high quantum yields to detect weak signals. Besides, fluorophores with various</div><div>emission frequencies are necessary to extract richer information. An essential com-</div><div>putational component to discover novel functional molecules is to predict molecular</div><div>properties. Here, we present statistical machines that predict excitation energies and</div><div>associated oscillator strengths of a given molecule using a random forest algorithm. Ex-</div><div>citation energies and oscillator strengths are directly related to the emission spectrum</div><div>and the quantum yields of fluorophores, respectively. We discovered specific molecu-</div><div>lar substructures and fragments that determine the oscillator strengths of molecules</div><div>from the feature importance analysis of our random forest machine. This discovery is</div><div>expected to serve as a new design principle for novel fluorophores.</div>

Deep Learning Optical Spectroscopy Based on Experimental Database: Potential Applications to Molecular Design

“… 38 , 39 Nakata and Shimazaki reported a large-scale electronic structure database (PubChemQC) 40 and recently, using the PubChemQC database, Lee and co-workers reported a random forest model to predict the highest oscillator strength and the corresponding excitation energy of molecules. 26 Lin and co-workers used a DL method to predict the bandgap of configurationally hybridized graphene and boron nitride, 21 and Chang and co-workers used a tuplewise material representation to predict the band gap of organic–inorganic perovskite, 2D materials, and binary and ternary inorganic materials. 27 Jiang and co-workers reported DL models that could predict infrared and ultraviolet absorption spectra from the conformations of a molecule using a theoretical database that was generated by molecular dynamics simulations and DFT calculations.…”

Machine Learning for Electronically Excited States of Molecules

“…In general, any type of descriptor might be suitable for a given problem. Applied descriptors range from topological and binary features generated from SMILES strings 533 to normal modes, which are often used as a coordinate system and descriptors to fit diabatic PESs (refs ( 16 , 99 , 136 , 143 , 145 , 145 − 147 , 149 , 392 , 534 )). Other types of molecular features besides structure-based ones, e.g., electronegativity, bond-order, oxidation states, ..., 17 , 71 are also used.…”

“…Recently, Kang et. al 533 used 500 000 molecules of the PubChemQC 578 database to train a random forest model on the excitation energy and the oscillator strength corresponding to the electronic state with the highest oscillator strength. Ten singlet states, as available in the PubChemQC database, were evaluated for that purpose.…”

“…The authors concluded that their study could serve the design of new fluorophores with high oscillator strengths. 533 …”

“…As can be seen, about half of the training sets are computed with multireference methods (refs ( 14 − 16 , 71 , 92 , 94 − 96 , 139 , 141 − 147 , 149 , 160 , 252 , 392 , 412 , 413 , and 660 )). The employed single-reference approaches are exclusively based on DFT (refs ( 17 , 77 , 93 , 97 , 98 , 152 , 158 , 358 , 510 , 533 , 539 , 547 , and 683 )). Analytical methods or experimental data are also applied.…”