Machine learning with quantum chemistry descriptors: predicting the solubility of small-molecule optoelectronic materials for organic solar cells

Abstract

A general solution prediction model was developed by using the smallest set of quantum chemistry descriptors.

Two-Dimensional Central Fused Ring Enabled High-Performance Quad-Rotor-Shaped Nonfullerene Acceptor Materials

“…The solubility of electron acceptor materials is a crucial factor that significantly impacts the morphology of the photoactive layer. Thus, we utilized a machine learning model to predict the solubility of Y6 and these 2D NFAs at 298 K. The molecular surface descriptors utilized for the machine learning model are summarized in Table S4. The predicted solubility of Y6 in chloroform is 43.8 mg mL –1 , showing excellent agreement with the experimental result (40 mg mL –1 ) .…”

“…The solubility of Y6, 2DC-ICH, 2DN-ICH, 2DO-ICH, and 2DS-ICH in 42 common solvents at 298 K was predicted using a machine learning model developed in our previous research. 21 This machine learning model is developed using the random forest regression algorithm and relies on a compact set of descriptors, comprising 7 bits of information. These descriptors include the area (molecular surface area), σ 2 + (positive ESP variance when analyzing the distribution of ESP on a vdW surface), and σ 2 − (negative ESP variance when analyzing the distribution of ESP on the vdW surface) of both the solvent and solute, along with temperature.…”

“…The solubility of Y6, 2DC-ICH, 2DN-ICH, 2DO-ICH, and 2DS-ICH in 42 common solvents at 298 K was predicted using a machine learning model developed in our previous research . This machine learning model is developed using the random forest regression algorithm and relies on a compact set of descriptors, comprising 7 bits of information.…”

Isomerization of two-dimensional non-fullerene electron acceptor materials for developing high-performance organic solar cells

“…The solubility of Y6, 2D-ICH, d-2D-ICH, and b-2D-ICH in 42 common solvents at 298 K was predicted by a machine learning model built in our previous work. 29 In this model, four 4-ethyloctane molecules were linked to each molecule of 2D-ICH, d-2D-ICH, and b-2D-ICH as side chains. This model is based on Random Forest regression algorithm with a small set of descriptors (7 bits), which consist of area (molecular surface area), s 2 + (positive ESP variance when analyzing the distribution of ESP on vdW surface) and s 2 -(negative ESP variance when analyzing the distribution of ESP on the vdW surface) of solvent and solute calculated by quantum chemistry and temperature.…”

“…Therefore, the solubility of electron acceptor materials should be considered during the design stage. The solubility of Y6, 2D-ICH, d-2D-ICH, and b-2D-ICH in 42 common solvents at 298 K was predicted by a machine learning model built in our previous work 29. As side chains can effectively affect the solubility of molecules, four 4-ethyloctane molecules were linked to each molecule of 2D-ICH, d-2D-ICH and b-2D-ICH, as shown in Fig.S2 (ESI †).…”