Molecular Property Prediction Based on Multi-Level Graph Self-Supervised Pretraining and Multi-Grain Finetuning.

Molecular property prediction is crucial for advancing medical research in areas like retrosynthesis analysis and drug discovery. The challenge of obtaining accurate molecular property labels has led to the use of pretrained Graph Neural Networks (GNNs) with self-supervised learning methods. However, traditional approaches often fail to capture detailed chemical structural and functional information, particularly within molecular functional groups, and do not adequately address relationships across molecular graph layers. To overcome these limitations, we introduce the Multi-level Molecular Graph Self-supervised Learning and Multi-grain Finetuning Framework (MMGSF). This framework consists of two components: Multi-level Molecular Graph Self-supervised Learning (MMGS) with a hierarchical GNN encoder to learn atom-motif-graph information and tailored pretraining tasks emphasizing inter-node relationships at various levels, and Multi-grain Finetuning (MGF) that refines node representations across grains, using a novel mol-adapter module with cross-attention for adaptive feature fusion. This fusion captures complex feature interactions, ensuring structural and semantic information from different grains contributes effectively to molecular property predictions. Superior results in molecular property classification tasks demonstrate the effectiveness of MMGSF, and its visualization performance shows that the learned representations capture chemical semantic information and properties successfully. This study offers fresh insights into the design of more effective self-supervised learning frameworks for molecular property prediction.