Sequence models
Convolutional and transformer-based models that map DNA sequence to chromatin state, RNA expression, and 3D contact.
Research
We develop computational models that bridge molecular measurements and biological function — from genome sequence to cell state to disease.
Genome regulation & cis-regulatory grammar
How does the non-coding genome encode when, where, and how much each gene is expressed? We train sequence-to-function deep-learning models on chromatin accessibility, histone modification, and transcription factor binding data to learn the regulatory grammar of enhancers and promoters — and to predict the functional consequences of non-coding variants relevant to human disease.
Variant effect prediction
In silico mutagenesis and interpretability methods that prioritize causal variants in GWAS loci and rare-disease cohorts.
Cross-species transfer
Comparative regulatory models that exploit evolutionary information to generalize across cell types, individuals, and species.
Single-cell & spatial multi-omics
Modern single-cell and spatial assays are extraordinarily rich — and extraordinarily sparse. We design principled probabilistic models and deep generative architectures that recover cell states, gene-regulatory programs, and tissue architecture from these noisy measurements, with calibrated uncertainty.
scATAC & scMultiome
Methods for footprinting, TF activity inference, and joint modeling of accessibility and expression at single-cell resolution.
Lineage & trajectory
Latent-variable and optimal-transport methods to reconstruct developmental and disease trajectories from snapshot data.
Spatial transcriptomics
Models that integrate spatial context with molecular profiles to map tissue niches and cell–cell signaling.
AI for human disease
Our methodological work is grounded in biological questions that matter clinically. In close collaboration with experimental and clinical groups at CUHK-Shenzhen and beyond, we apply these models to cancer, immune, and metabolic disease — with an emphasis on interpretability, robustness, and reproducibility.
Cancer genomics
Decoding non-coding driver variation and tumor microenvironment heterogeneity from bulk and single-cell tumor profiles.
Immune dynamics
Modeling T- and B-cell receptor repertoires and immune cell states across health and disease.
Foundation models
Building and adapting biological foundation models so that small clinical cohorts benefit from large reference atlases.