Jianbo Qiao

A Token-Pruning Framework Enables Efficient Representation of the Human Genome for RNA Modification Analysis.

Relation-Aware Pretraining and Reaction Center Modeling for Chemical Reaction Graph Representation Learning.

Learning informative representations of chemical reactions is crucial for advancing synthesis-related downstream tasks, such as reaction condition prediction and yield prediction. However, existing models often struggle to simultaneously capture the global reactant-product relationships and the localized structural evolutions at the reaction center within a unified framework. To bridge this gap, we propose the Knowledge-Aware Graph Transformer (KAGT), a framework for chemical reaction graph representation learning. KAGT combines a shared graph Transformer encoder with a condition-attributed reaction knowledge graph, in which reactant and product molecular graphs serve as source and target entities and solvent, catalyst, and temperature descriptors define latent relation attributes for relation-aware pretraining. The pretraining strategy further includes masked atom modeling and 3D geometric denoising based on computed molecular conformers. We also introduce an explicit reaction-center modeling mechanism guided by atom mapping to capture fine-grained local transformations. Evaluations across three downstream tasks show that KAGT achieves strong performance relative to existing baselines in reaction classification, reaction condition prediction, and yield prediction. Qualitative case studies further show that the learned center scores concentrate on chemically plausible transformation sites, supporting KAGT as a transferable representation framework for AI-driven chemical synthesis.

Hyp-retro: hypernetwork-guided preference-conditioned retrosynthetic planner.

PeptideSGCL: Structure-Enhanced Graph-Transformer Encoding and Dual-Level Contrastive Learning for Peptide Property Prediction.

Peptides play important roles in biological processes and biomedical applications, and their hemolytic (Hemo) and nonfouling (NF) properties directly affect their safety and translational potential. Therefore, accurate predictive models are essential for the rational design of functional peptides. Although existing multimodal peptide property prediction methods can jointly exploit sequence and structural information, their structural encoders still rely primarily on local graph convolution and their contrastive objectives are largely focused on cross-modal alignment. Consequently, they remain limited in modeling long-range structural dependencies and in enhancing intramodal discriminability. To address these limitations, we propose a multimodal dual-contrastive learning framework for peptide property prediction, which improves both the structural encoder and the contrastive learning strategy to enhance the quality of joint sequence-structure representations. Specifically, ProtBERT is adopted as the sequence encoder, and a hierarchical GNN-Transformer structural encoder is constructed to capture local topological patterns and long-range structural dependencies. In addition, a parallel graph spatial channel attention module is introduced to enhance task-relevant structural features. Within a shared embedding space, we further design an interintra hybrid supervised contrastive learning strategy to jointly optimize sequence-structure alignment and intramodal class discriminability. Experimental results show that the proposed method achieves overall performance superior to baseline models on both hemolysis and NF prediction tasks, providing an effective framework for multimodal representation learning in peptide-property prediction.