Marine natural products (MNPs) are a diverse group of bioactive compounds with varied chemical structures, but their biological origins are often misannotated due to complex host–microbe symbiosis. Propagated through public databases, such errors hinder biosynthetic studies and AI-driven drug discovery. Here, we develop a structure-based workflow of origin classification and misannotation correction for marine datasets. Using CMNPD and NPAtlas compounds, we integrate a two-step cleaning strategy that detects label inconsistencies and filters structural outliers with a microbial-pretrained graph neural network. The optimized model achieves a balanced accuracy of 85.56% and identifies 3996 compounds whose predicted microbial origins contradict their Animalia labels. These putative symbiotic metabolites cluster within known high-risk taxa, and interpretability analysis reveal biologically coherent structural patterns. This framework provides a scalable quality-control approach for natural product databases and supports more accurate biosynthetic gene cluster (BGC) tracing, host selection, and AI-driven marine natural product discovery.

Inter-tissue communication (ITC) is critical for maintaining the physiological functions of multiple tissues and is closely related to the onset and development of various complex diseases. Nevertheless, there is no well-organized data resource for known ITC molecules with explicit ITC routes from source tissues to target tissues. To address this issue, in this work, we manually reviewed nearly 190,000 publications and identified 1408 experimentally supported ITC entries in which the ITC molecules, their communication routes, and their functional annotations were included. To facilitate our work, these curated ITC entries were incorporated into a user-friendly database named IntiCom-DB. This database also enables visualization of the expression abundances of ITC proteins and their interaction partners. Finally, bioinformatics analyses on these data revealed common biological characteristics of the ITC molecules. For example, tissue specificity scores of ITC molecules at the protein level are often higher than those at the mRNA level in the target tissues. Moreover, the ITC molecules and their interaction partners are more abundant in both the source tissues and the target tissues. IntiCom-DB is freely available as an online database. As the first comprehensive database of ITC molecules with explicit ITC routes to the best of our knowledge, we hope that IntiCom-DB will benefit future ITC-related studies.

Chemical reactions are the fundamental building blocks of drug design and organic chemistry research. In recent years, there has been a growing need for a large-scale deep-learning framework that can efficiently capture the basic rules of chemical reactions. In this paper, we have proposed a unified framework that addresses both the reaction representation learning and molecule generation tasks, which allows for a more holistic approach. Inspired by the organic chemistry mechanism, we develop a novel pretraining framework that enables us to incorporate inductive biases into the model. Our framework achieves state-of-the-art results on challenging downstream tasks. By possessing chemical knowledge, our generative framework overcome the limitations of current molecule generation models that rely on a small number of reaction templates. In the extensive experiments, our model generates synthesizable drug-like structures of high quality. Overall, our work presents a significant step toward a large-scale deep-learning framework for a variety of reaction-based applications.

Emerging artificial intelligence (AI)-based genome mining methods have revolutionized the paradigm of bacterial secondary metabolite (SM) discovery. Fortunately, recent data accumulation of fungal biosynthetic gene clusters (BGC) fairly offers opportunities for systematic development and evaluation of fungi-specific pipelines. In this work, we proposed a deep learning framework termed as f-BGM specifically for fungal genome mining. By designing a novel self-attention-based architecture to augment inter-domain associations in local genomic contexts, f-BGM exhibits superior performance over existing AI-based methods in both in-distribution and out-of-distribution benchmark tests for BGC detection. Further analyses demonstrate that f-BGM is of decent interpretability on deciphering single-domain and -protein importance, as well as inter-domain partnership. By establishing additional binary classification models, f-BGM also achieves high-quality identification of core enzymes within given BGCs. Finally, case studies of f-BGM-driven genome mining in marine fungi uncovers biosynthetic potential underestimated by the rule-based method antiSMASH, as supported by experimental and computational validation.