Hongru Hu’s research while affiliated with University of California System and other places

A bstract Single-cell RNA sequencing (scRNA-seq) enables high-resolution exploration of cellular diversity and gene regulation, yet analyzing such data remains challenging due to technical and methodological limitations. Existing task-specific deep generative models like Variational Auto-Encoder (VAE) and its variants struggle to incorporate external biological knowledge, while transformer-based foundational large Language Models (LLMs or large LaMs) face limitations in computational cost and applicability to tabular gene expression data. Here, we introduce sciLaMA (single-cell interpretable Language Model Adapter), a novel representation learning framework that bridges these gaps by integrating static gene embeddings from multimodal LaMs with scRNA-seq tabular data through a paired-VAE architecture. Our approach generates context-aware representations for both cells and genes and outperforms state-of-the-art methods in key single-cell downstream tasks, including batch effect correction, cell clustering, and cell-state-specific gene marker and module identification, while maintaining computational efficiency. sciLaMA offers a computationally efficient, unified framework for comprehensive single-cell data analysis and biologically interpretable gene module discovery.

Multimodal single-cell assays profile multiple sets of features in the same cells and are widely used for identifying and mapping cell states between chromatin and mRNA and linking regulatory elements to target genes. However, the high dimensionality of input features and shallow sequencing depth compared to unimodal assays pose challenges in data analysis. Here we present scPair, a multimodal single-cell data framework that overcomes these challenges by employing an implicit feature selection approach. scPair uses dual encoder-decoder structures trained on paired data to align cell states across modalities and predict features from one modality to another. We demonstrate that scPair outperforms existing methods in accuracy and execution time, and facilitates downstream tasks such as trajectory inference. We further show scPair can augment smaller multimodal datasets with larger unimodal atlases to increase statistical power to identify groups of transcription factors active during different stages of neural differentiation.

Multi-modal single cell RNA assays capture RNA content as well as other data modalities, such as spatial cell position or the electrophysiological properties of cells. Compared to dedicated scRNA-seq assays however, they may unintentionally capture RNA from multiple adjacent cells, exhibit lower RNA sequencing depth compared to scRNA-seq, or lack genome-wide RNA measurements. We present scProjection, a method for mapping individual multi-modal RNA measurements to deeply sequenced scRNA-seq atlases to extract cell type-specific, single cell gene expression profiles. We demonstrate several use cases of scProjection, including identifying spatial motifs from spatial transcriptome assays, distinguishing RNA contributions from neighboring cells in both spatial and multi-modal single cell assays, and imputing expression measurements of un-measured genes from gene markers. scProjection therefore combines the advantages of both multi-modal and scRNA-seq assays to yield precise multi-modal measurements of single cells.

Multimodal assays such as Patch-seq that simultaneously profile molecular and cellular phenotypes of cells enable the identification of molecular underpinnings of electrophysiological response patterns in neurons. Here we analyzed Patch-seq measurements of thousands of mouse interneurons to identify subclass-specific genes associated with different electrophysiological features. We found extensive subclass specificity: even for the same ephys feature, largely unique sets of genes are associated with that feature in different subclasses. Well established ephys genes such as Reln demonstrated subclass specificity that was previously not reported. Surprisingly, we found that ion channels explained significantly less variation in ephys response across interneurons compared to other genes; in particular, gene sets enriched in cell adhesion genes were amongst the most associated. We found our gene sets associated with action potential dV/dt measurements explained significant heritability of Schizophrenia risk, suggesting a novel role of single neuron electrophysiology in Schizophrenia risk. Finally, we observed significant ephys function switching of cell adhesion molecules across subclasses; the same adhesion molecule was observed to associate with different functional ephys measurements in distinct subclasses and co-express with different genes, suggesting re-purposing of adhesion molecules in different subclasses. Overall, our results yield novel insight into the specificity of roles that individual genes and adhesion molecules play in both single neuron ephys response and Schizophrenia risk.