作者: Qingyou Zhang , Fangfang Zheng , Tanfeng Zhao , Xiaohui Qu , João Aires-de-Sousa
关键词: Rank (linear algebra) 、 Fukui function 、 Kernel (statistics) 、 Artificial intelligence 、 Quantum chemistry 、 Chemistry 、 Machine learning 、 Organic molecules 、 Atom 、 Molecule 、 Quantitative structure–activity relationship
摘要: To enable the fast estimation of atom condensed Fukui functions, machine learning algorithms were trained with databases DFT pre-calculated values for ca. 23,000 atoms in organic molecules. The problem was approached as ranking types Bradley-Terry (BT) model, and regression function. Random Forests (RF) to predict function, rank a molecule, classify high/low Atomic descriptors based on counts spheres around kernel atom. BT coefficients assigned enabled identification (93-94 % accuracy) highest function pairs same molecule differences ≥0.1. In whole molecules, top could be recognized 50 % cases and, average, about 3 4 shortlist 4. Regression RF yielded predictions test sets R(2) =0.68-0.69, improving ability molecule. Atom classification (as function) obtained sensitivity 55-61 % specificity 94-95 %.