MF-SuP-pK_(a): Multi-fidelity modeling with subgraph pooling mechanism for pK_(a) prediction  

作　　者：Jialu Wu Yue Wan Zhenxing Wu Shengyu Zhang Dongsheng Cao Chang-Yu Hsieh Tingjun Hou 

机构地区：[1]Innovation Institute for Artificial Intelligence in Medicine of Zhejiang University,College of Pharmaceutical Sciences,Zhejiang University,Hangzhou 310058,China [2]CarbonSilicon AI Technology Co.,Ltd.,Hangzhou 310018,China [3]Tencent Quantum Laboratory,Tencent,Shenzhen 518057,China [4]Xiangya School of Pharmaceutical Sciences,Central South University,Changsha 410004,China

出　　处：《Acta Pharmaceutica Sinica B》2023年第6期2572-2584,共13页药学学报（英文版）

基　　金：financially supported by National Key Research and Development Program of China (2021YFF1201400);National Natural Science Foundation of China (22220102001);Natural Science Foundation of Zhejiang Province (LZ19H300001, LD22H300001, China)。

摘　　要：Acid-base dissociation constant(pK_(a)) is a key physicochemical parameter in chemical science, especially in organic synthesis and drug discovery. Current methodologies for pK_(a) prediction still suffer from limited applicability domain and lack of chemical insight. Here we present MF-SuP-pK_(a)(multi-fidelity modeling with subgraph pooling for pK_(a) prediction), a novel pK_(a) prediction model that utilizes subgraph pooling, multi-fidelity learning and data augmentation. In our model, a knowledgeaware subgraph pooling strategy was designed to capture the local and global environments around the ionization sites for micro-pK_(a) prediction. To overcome the scarcity of accurate pK_(a) data, lowfidelity data(computational pK_(a)) was used to fit the high-fidelity data(experimental pK_(a)) through transfer learning. The final MF-SuP-pK_(a) model was constructed by pre-training on the augmented ChEMBL data set and fine-tuning on the DataWarrior data set. Extensive evaluation on the DataWarrior data set and three benchmark data sets shows that MF-SuP-pK_(a) achieves superior performances to the state-of-theart pK_(a) prediction models while requires much less high-fidelity training data. Compared with Attentive FP, MF-SuP-pK_(a) achieves 23.83% and 20.12% improvement in terms of mean absolute error(MAE) on the acidic and basic sets, respectively.

关 键 词：pK_(a)prediction Graph neural network Subgraph pooling Multi-fidelity learning Data augmentation 

分 类 号：O64[理学—物理化学]