Zhang, Shuwen

Abstract Motivation Single-cell RNA sequencing (scRNA-seq) has become an important method for characterizing cellular heterogeneity, revealing more biological insights than the bulk RNA-seq. The surge in scRNA-seq data across multiple individuals calls for efficient and statistically powerful methods for differential expression (DE) analysis that addresses individual-level biological variability. Results We introduced DiSC, a method for conducting individual-level DE analysis by extracting multiple distributional characteristics, jointly testing their association with a variable of interest, and using a flexible permutation testing framework to control the false discovery rate (FDR). Our simulation studies demonstrated that DiSC effectively controlled the FDR across various settings and exhibited high statistical power in detecting different types of gene expression changes. Moreover, DiSC is computationally efficient and scalable to the rapidly increasing sample sizes in scRNA-seq studies. When applying DiSC to identify DE genes potentially associated with COVID-19 severity and Alzheimer’s disease across various types of peripheral blood mononuclear cells and neural cells, we found that our method was approximately 100 times faster than other state-of-the-art methods and the results were consistent and supported by existing literature. While DiSC was developed for scRNA-seq data, its robust testing framework can also be applied to other types of single-cell data. We applied DiSC to cytometry by time-of-flight data, DiSC identified significantly more DE markers than traditional methods. Availability and implementation The R software package “SingleCellStat” is freely available on CRAN (https://cran.r-project.org/web/packages/SingleCellStat/index.html) and GitHub (https://github.com/Lujun995/DiSC). The replication code for reproducing the analyses in this study is publicly accessible at https://github.com/Lujun995/DiSC_Replication_Code. The scRNA-seq expression matrix and metadata utilized in our simulations and analyses can be retrieved from https://cells.ucsc.edu/autism/rawMatrix.zip, https://cellxgene.cziscience.com/collections/1ca90a2d-2943-483d-b678-b809bf464c30, and https://covid19.cog.sanger.ac.uk/submissions/release1/haniffa21.processed.h5ad.

Abstract 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 sciL-aMA (single-cell interpretable Language Model Adapter), a novel representation learning frame-work that bridges these gaps by integrating static gene embeddings from multimodal LLMs 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. sciL-aMA offers a computationally efficient, unified framework for comprehensive single-cell data analysis and biologically interpretable gene module discovery. Source code is available at https://github.com/microsoft/sciLaMA