Dr. Liu’s research lab develops computational methods, tools, and resources for understanding the human genome and diseases such as diabetes. Recently, the methodology focuses are knowledge graphs and foundation models. We have developed a knowledge graph GenomicKB to accumulate human-readable knowledge about the human genome. We have extracted genomic knowledge from PubMed and developed another knowledge graph GLKB. We have also developed a genomic foundation model EPCOT which comprehensively predicts multiple genomic modalities.
The lab currently participates in several NIH consortia, including 4DN, IGVF, HIRN, CFDE, and the recent PanKbase program. In particular, Dr. Liu co-leads the Machine Learning Focus Group at IGVF, co-leads the Data/Metadata Working Group at PanKbase, and leads the development of PanKgraph, the knowledge graph within the PanKbase system.
PhD in Computer Science, 2014
University of Wisconsin - Madison
10/2024 Dr. Liu presented “Building knowledge graphs towards transparent biomedical AI” at NIH/NIDDK AI in Precision Medicine Workshop link
08/2024 Dr. Liu received Endowment for the Basic Sciences Teaching Award 2024. link
08/2024 Yicheng Tao’s CNTools work was published on PLoS CB.
04/2024 Shuze’s work was published on Developmental Cell. link
A graph database for human genome, epigenome, transcriptome, and 4D nucleome.
A computational tool for identifying nucleosomes in ultra-high resolution contact maps.
A deep learning model for characterizing collaborative transcription regulation.
A generalizable framework to comprehensively predict epigenome, chromatin organization, and transcriptome.
A deep learning model for connecting high-resolution 3D chromatin organization with epigenomics.
An unsupervised manifold alignment algorithm, MMD-MA, for integrating multiple measurements carried out on disjoint aliquots of a given population of single cells.
The first computational embedding method for single cells in terms of their chromatin organization.
A graphical model based multiple testing procedure which captures dependence among the hypotheses.