The interrogation of cellular states and interactions in immunology research is an ever-evolving task, requiring adaptation to the current levels of high dimensionality. Cytometry enables high-dimensional profiling of immune cells, but its analysis is hindered by the complexity and variability of the data. We present MAESTRO, a self-supervised set representation learning model that generates vector representations of set-structured data, which we apply to learn immune profiles from cytometry data. Unlike previous studies only learn cell-level representations, whereas MAESTRO uses all of a sample's cells to learn a set representation. MAESTRO leverages specialized attention mechanisms to handle sets of variable number of cells and ensure permutation invariance, coupled with an online tokenizer by self-distillation framework. We benchmarked our model against existing cytometry approaches and other existing machine learning methods that have never been applied in cytometry. Our model outperforms existing approaches in retrieving cell-type proportions and capturing clinically relevant features for downstream tasks such as disease diagnosis and immune cell profiling.

MOTIVATION: statistics from genome-wide association studies (GWAS) are widely used in fine-mapping and colocalization analyses to identify causal variants and their enrichment in functional contexts, such as affected cell types and genomic features. With the expansion of functional genomic (FG) datasets, which now include hundreds of thousands of tracks across various cell and tissue types, it is critical to establish scalable algorithms integrating thousands of diverse FG annotations with GWAS results. RESULTS: We propose BTS (Bayesian Tissue Score), a novel, highly efficient algorithm uniquely designed for (i) identifying affected cell types and functional elements (context-mapping) and (ii) fine-mapping potentially causal variants in a context-specific manner using large collections of cell type-specific FG annotation tracks. BTS leverages GWAS summary statistics and annotation-specific Bayesian models to analyze genome-wide annotation tracks, including enhancers, open chromatin, and histone marks. We evaluated BTS on GWAS summary statistics for immune and cardiovascular traits, such as Inflammatory Bowel Disease (IBD), Rheumatoid Arthritis (RA), Systemic Lupus Erythematosus (SLE), and Coronary Artery Disease (CAD). Our results demonstrate that BTS is over 100× more efficient in estimating functional annotation effects and context-specific variant fine-mapping compared to existing methods. Importantly, this large-scale Bayesian approach prioritizes both known and novel annotations, cell types, genomic regions, and variants and provides valuable biological insights into the functional contexts of these diseases. AVAILABILITY AND IMPLEMENTATION: Docker image is available at https://hub.docker.com/r/wanglab/bts with preinstalled BTS R package (https://bitbucket.org/wanglab-upenn/BTS-R) and BTS GWAS summary statistics analysis pipeline (https://bitbucket.org/wanglab-upenn/bts-pipeline).

MOTIVATION: Cell-type deconvolution of bulk tissue RNA sequencing (RNA-seq) data is an important step toward understanding the variations in cell-type composition among disease conditions. Owing to recent advances in single-cell RNA sequencing (scRNA-seq) and the availability of large amounts of bulk RNA-seq data in disease-relevant tissues, various deconvolution methods have been developed. However, the performance of existing methods heavily relies on the quality of information provided by external data sources, such as the selection of scRNA-seq data as a reference and prior biological information. RESULTS: We present the Integrated and Robust Deconvolution (InteRD) algorithm to infer cell-type proportions from target bulk RNA-seq data. Owing to the innovative use of penalized regression with a new evaluation criterion for deconvolution, InteRD has three primary advantages. First, it is able to effectively integrate deconvolution results from multiple scRNA-seq datasets. Second, InteRD calibrates estimates from reference-based deconvolution by taking into account extra biological information as priors. Third, the proposed algorithm is robust to inaccurate external information imposed in the deconvolution system. Extensive numerical evaluations and real-data applications demonstrate that InteRD yields more accurate and robust cell-type proportion estimates that agree well with known biology. AVAILABILITY AND IMPLEMENTATION: The proposed InteRD framework is implemented in R and the package is available at https://cran.r-project.org/web/packages/InteRD/index.html. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.