Transformers for molecular property prediction: domain adaptation efficiently improves performance.

This comparison reveals an important trend: the top-performing representation in this study (RDKit descriptors, MolBERT, and MLM_MTR) incorporate explicit physicochemical and 2D information.

also in 625 other papers

The results show that domain adaptation using the MTR objective consistently yields highly significant improvements across all datasets (P-value < \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$4e^{-9}$$\end{document} 4 e - 9 ).

also in 83,176 other papers