Featured Publications

Early organogenesis is a crucial stage in embryonic development, characterized by extensive cell fate specification to initiate organ formation but also by a high susceptibility to developmental defects. Here, we profiled 285 serial sections from six E7.5–E8.0 embryos to generate full spatiotemporal transcriptome and signal maps during early organogenesis at single-cell resolution. By developing SEU-3D, we reconstructed digital embryos, enabling investigation of regionalized gene expression in the native spatial context. We established a space-informed gene-cell co-embedding approach, systematically characterized the spatial atlas of endoderm and mesoderm derivatives, and elucidated signaling networks across germ layers and cell types.

In this paper, we consider deep approximate policy iteration (DAPI) with the Bellman residual minimization in reinforcement learning. In each iteration of DAPI, we apply convolutional neural networks (CNNs) with ReLU activation, called ReLU CNNs, to estimate the fixed point of the Bellman equation by minimizing an unbiased minimax loss.

We present SDEvelo, a generative approach to inferring RNA velocity by modeling the dynamics of unspliced and spliced RNAs via multivariate stochastic differential equations (SDE). Uniquely, SDEvelo explicitly models inherent uncertainty in transcriptional dynamics while estimating a cell-specific latent time across genes.

We propose X-ING (Cross-INtegrative Genomics) for cross-omics and cross-context integrative analysis.

We propose the use of a unified and principled probabilistic model, PRECAST, to simultaneously estimate low-dimensional embeddings for biological effects between cell/domain types, perform spatial clustering, and most importantly, align embeddings for normalized gene expression matrices from multiple tissue slides.

We propose the use of a probabilistic model, SpatialAnno, which performs cell/domain-type assignments for SRT data and has the capability of leveraging non-marker genes to assign cell/domain types via a factor model while accounting for spatial information via a Potts model.

We propose a unified and principled method to both estimate low-dimensional embeddings relevant to latent class labels and, in the case of spatial transcriptomics analysis, further leverage these embeddings with spatial information to perform spatial clustering using an HMRF.

We propose a tissue-specific collaborative mixed model (TisCoMM) for TWASs, leveraging the co-regulation of genetic variations across different tissues explicitly via a unified probabilistic model.