Semantic Knowledge Mining: Methods and Applications

Gene functional similarity calculations mathematically quantify biomedical knowledge to gauge the similarity between genes. This forms the basis for solving various challenges, such as predicting gene attributes and analyzing diseases and drugs. To facilitate this process, we developed the software tool GOSemSim, built upon gene ontology principles. This work was published in Bioinformatics in 2010. We have extended this work to encompass diverse biomedical knowledge domains, including diseases, phenotypes, and medical subject headings.

To identify biological pathways associated with perturbed genes, we developed a suite of R packages, spearheaded by clusterProfiler. This tool facilitates online retrieval of genome annotations, supporting enrichment analysis for thousands of species. It also provides a universal interface for user-provided custom annotations. Notably, clusterProfiler supports the utilization of genomic regions for enrichment analysis. We have extended its application to disease ontology, medical subject headings, and Reactome pathway analysis. This work has been widely cited and integrated into numerous bioinformatics software tools.

To address complex experimental designs involving multiple time points and conditions, we developed clusterProfiler for comparing biological themes. This contribution was published in OMICS: A Journal of Integrative Biology in 2012. The tool enables users to specify complex experimental designs using formula syntax, allowing for the comparison of functional profiles across different conditions. We authored a protocol article in Nature Protocols in 2024, detailing the application of this comparative framework.

We developed the ChIPseeker package to bridge the gap between genomic cis-regulatory positions and biological function. It facilitates annotation and comparison of genomic peaks, integrating with the GEO database for data mining. This work was published in Bioinformatics in 2015. ChIPseeker is widely used for analyzing various cistromic datasets, including ChIP-seq, DNase-seq, and ATAC-seq.

To enhance the interpretability of single-cell clustering, we proposed integrating biological knowledge as attributes of graph nodes. We developed the MSGNN tool based on graph neural networks to realize this advancement, improving the biological interpretation of complex data.

We proposed a method for characterizing cell functional states and identifying spatially highly variable features. We developed the SVP package to facilitate these methods, enabling the characterization of biological functions and the determination of their spatial distribution specificity.

Users familiar with R/Bioconductor have several R packages to perform epigenomic enrichment analysis. ChIPseeker (Yu et al., 2015) provides a variety of functions for annotating user-provided ROIs in BED format by proximity to the nearby genes, finding overlaps with epigenomic regions, performing gene-centric and epigenomic enrichment analysis. ChIPseeker employs permutation strategy to calculate overlap enrichment p-values. ChIPseeker provides access to multi-organism epigenomic datasets from GEO (Gene Expression Omnibus), expanding enrichment analysis to species other than human. Coupled with excellent visualization capabilities, ChIPseeker is a well-designed tool with complete functionality for the interpretation of genome-wide ROIs.

– Excerpt from the article: Epigenomic annotation-based interpretation of genomic data: from enrichment analysis to machine learning, Bioinformatics, 2017, 33(20):3323-3330.