Gene expression is a complex process modulated by the interplay of proximal and distal regulatory domains in 3D space. These domains, transcription factors (TFs), and target genes form regulatory circuits. TF binding to chromatin regions and 3D looping between these regions and gene promoters are key mechanisms governing how TFs translate regulatory signals into changes in RNA transcription. In disease, these circuits can be dysregulated in a cell-type-specific manner, often masked in bulk sample analyses. Therefore, identifying the impact of disease on regulatory circuits requires a framework capable of mapping regulatory domains with chromatin accessibility changes to altered gene expression at cell-type resolution. Single-cell RNA sequencing (scRNA-seq) and single-cell transposase-accessible chromatin sequencing (scATAC-seq) have improved the identification of differential chromatin sites and differentially expressed genes within individual cell types. However, advances in single-cell assay technology have outpaced the development of methods to fully utilize multiomics datasets for studying disease-associated regulation, particularly for regulatory interactions not directly measured. Existing computational approaches show promise but lack the capacity to resolve regulatory changes within individual cell types, hindering the elucidation of disease-affected regulatory circuits and their varying responses across disease states. This paper addresses these limitations by introducing MAGICAL, a novel method.

The paper reviews existing methods for analyzing single-cell multiomics data, highlighting their limitations in resolving cell-type-specific regulatory circuits. It mentions several previous computational approaches that attempt to integrate scRNA-seq and scATAC-seq data, but these methods either fail to achieve cell-type resolution or lack the capacity to model the complex interplay between chromatin accessibility and gene expression changes in different disease conditions. The authors emphasize the need for a method that can accurately identify disease-associated regulatory circuits at the cell-type level and differentiate between varying disease states, which is crucial for understanding disease mechanisms and developing targeted therapies. The review implicitly supports the need for MAGICAL by pointing to the gap in current methodologies.

MAGICAL (Multiome Accessibility Gene Integration Calling and Looping) is a hierarchical Bayesian approach designed to identify disease-associated regulatory circuits from paired single-cell RNA sequencing (scRNA-seq) and single-cell Assay for Transposase-Accessible Chromatin using sequencing (scATAC-seq) data. The framework incorporates TF motifs and Topologically Associated Domain (TAD) boundaries as prior information. First, differentially accessible sites (DAS) within each cell type are associated with TFs via motif sequence matching and linked to differentially expressed genes (DEG) by genomic localization within the same TAD. MAGICAL uses a Bayesian framework to model chromatin accessibility and gene expression variation across cells and samples in each cell type, estimating the confidence of TF-peak and peak-gene linkages for each candidate circuit. To accurately detect differences in regulatory circuit activity, MAGICAL explicitly models signal and noise in both data types. A TF-peak binding variable and a hidden TF activity variable are jointly estimated to fit chromatin accessibility variation, and these variables are used with a peak-gene looping variable to fit gene expression variation. Gibbs sampling iteratively estimates variable values, optimizing circuit TF-peak-gene linkages. High-confidence circuits fitting signal variation in both data types are selected. Importantly, MAGICAL models TF activity separately from TF expression, allowing analysis of single-cell true multiome or sample-paired multiomics datasets. The method's accuracy is validated through benchmarking against multiple public datasets, including comparisons with existing integrative methods like TRIPOD and FigR, demonstrating superior performance in inferring regulatory circuits.

MAGICAL accurately identifies disease-associated regulatory circuits by integrating scRNA-seq and scATAC-seq data. In a study of *Staphylococcus aureus* sepsis, MAGICAL identified sepsis-associated regulatory circuits predominantly in CD14 monocytes. The method successfully distinguished between responses to methicillin-resistant (*MRSA*) and methicillin-susceptible (*MSSA*) *S. aureus* infections, identifying epigenetic circuit biomarkers that differentiate the two, even when differential expression analysis failed. In a COVID-19 study, MAGICAL identified high-confidence circuits across different cell types (CD8 effector memory T cells, CD14 monocytes, and natural killer cells) for both mild and severe clinical groups. The precision of MAGICAL-selected chromatin sites and genes was significantly higher than that of traditional methods. For *S. aureus*, MAGICAL identified 1,513 high-confidence regulatory circuits across 13 major cell types. Circuit genes in CD14 monocytes showed significant enrichment of AP-1 complex proteins as key regulators. MAGICAL accurately identified distal regulatory chromatin sites (15-25 kb from TSS), many overlapping with enhancer-like regions and GWAS loci associated with inflammation. Circuit genes were significantly enriched with epigenetically driven genes. Crucially, MAGICAL-identified circuit genes robustly predicted *S. aureus* infection in multiple independent datasets and showed predictive value for antibiotic sensitivity, unlike traditional differential expression analysis. The consistent performance across different cohorts highlighted the fundamental role of these regulatory processes in the host response to *S. aureus* sepsis.

MAGICAL addresses a critical unmet need: identifying differential regulatory circuits from single-cell multiomics data of contrasting conditions, including distal chromatin sites. This capability is particularly important, as distal regulatory sites are increasingly recognized as crucial for gene regulatory mechanisms. While MAGICAL excels at cell-type-specific analysis, it struggles with poorly defined cell types or conditions due to a limited number of candidate peaks and genes. Future work could focus on improving circuit identification in such cases and explicitly modeling cell-type specificity in disease circuit identification. The success of MAGICAL in predicting *S. aureus* infection and antibiotic sensitivity demonstrates its potential for improving disease diagnosis and treatment. The ability to identify regulatory differences, even between closely related bacterial infections like MRSA and MSSA, showcases MAGICAL's power in uncovering subtle yet significant variations in the host immune response. This work provides a significant advance in bioinformatics tools for interpreting single-cell multiomics data and has broad implications for various disease research areas.

MAGICAL is a novel, accurate method for mapping disease-associated regulatory circuits from single-cell multiomics data. It addresses limitations of existing methods by achieving cell-type resolution and identifying regulatory circuits involving distal chromatin sites. Its success in predicting *S. aureus* infection and antibiotic sensitivity highlights its potential for translational applications. Future directions include improving its performance with poorly defined cell types and directly modeling cell-type specificity in disease circuit identification.

MAGICAL's performance can be affected by the quality and definition of cell types in the input data. For less distinct cell types or conditions, the inference of cell-type-resolved circuits becomes challenging due to a limited number of candidate peaks and genes. The method's reliance on prior information, such as TF motifs and TAD boundaries, might introduce biases. Additionally, the cell-type specificity of disease circuits is not directly modeled, but rather inferred from separate analysis of each cell type.