Saurabh Sinha

Motivation: Joint Spatial Metabolomics (SM) and Spatial Transcriptomics (ST) profiling is a powerful approach to fine-mapping of metabolic states associated with tissue function. Current computational tools for analysis of “SM+ST” data focus primarily on alignment and integration of the two modalities, with limited support for probing biological relationships between the two molecular layers. Results: We present MANTIS, a statistical framework for analyzing co-registered SM+ST profiles at single cell or spot resolution, along with spatial domain or cell type information, to discover metabolite spatial patterns and gene-metabolite relationships. It employs an autocorrelation-preserving permutation strategy to assess statistical significance, yielding calibrated inference under spatial dependence. It disentangles different sources of spatial patterns and correlations, viz., those arising from regional preferences, cell type associations, or other unknown factors. It introduces the use of spatial cross-correlation and spatial partial correlation statistics for quantifying gene-metabolite associations. Across data sets spanning different spatial technologies, tissues and species, MANTIS provides more specific and interpretable discoveries than existing methods through rigorous statistical testing and explicitly modeling confounding structure. To our knowledge, MANTIS is the first toolkit to unify spatial metabolomics, spatial transcriptomics, cell type information and spatial domains within a single framework that emphasizes spatial statistics, hypothesis testing and confounder correction. Availability and Implementation: Freely available on the web at https://github.com/yuhaotuo/MANTIS .

Breast and other solid tumors respond poorly to immune therapy. Myeloid cells (MCs) such as macrophages contribute to resistance. Established clinical evidence links cholesterol to cancer outcomes, with MC function being regulated by cholesterol metabolism. We screened MC-expressed regulators of cholesterol homeostasis linked to survival and identified the cholesterol efflux protein ABCA1. ABCA1 activity increases anticancer functions of macrophages: enhancing tumor infiltration, decreasing angiogenic potential, reducing efferocytosis, and improving support of CD8+ T cell activity. Mechanistically, different AKT isoforms are involved, through both PI3K-dependent and PI3K-independent mechanisms. Highlighting the clinical relevance of our findings are correlations between ABCA1 in macrophages and angiogenic potential, VEGFA , and CD8 T cell abundance and activity. The culmination of these activities was demonstrated through increased tumor growth and metastasis in mice lacking MC-expressed ABCA1. Tumors grown in these mice were also more resistant to immune therapy. Therefore, modulating ABCA1 activity within MCs may represent a previously unidentified approach to immune therapy.

Spatial transcriptomics has enabled the study of mRNA distributions within cells, a key aspect of cellular function. However, there is a dearth of tools that can identify and interpret functionally relevant spatial patterns of subcellular transcript distribution. To address this, we present CellSP, a computational framework for identifying, visualizing, and characterizing consistent subcellular spatial patterns of mRNA. CellSP introduces the concept of “gene-cell modules”, which are gene sets with coordinated subcellular transcript distributions in many cells. It provides intuitive visualizations of the captured patterns and offers functional insights into each discovered module. We demonstrate that CellSP reliably identifies functionally significant modules across diverse tissues and technologies. We use the tool to discover subcellular spatial phenomena related to myelination, axonogenesis, and synapse formation in the mouse brain. We find immune response-related modules that change between kidney cancer and healthy samples, and myelination-related modules specific to mouse models of Alzheimer’s Disease.

Metabolic dysregulation is an important factor for osteoarthritis pathogenesis, but comprehensive studies of underlying mechanisms and pathways are rare. We analyzed newly generated metabolomics data on bone marrow from 119 osteoarthritis patients, along with single-cell transcriptomics data to reconstruct networks of gene-metabolite associations at cell type resolution. Hubs of these networks – cell type-specific as well as pan-cell type hubs – revealed key molecular factors of osteoarthritis heterogeneity. Systems-level analysis of hubs revealed major roles for glycerophospholipid, glycerolipid and sphingolipid metabolism pathways, along with lipid signaling. We used Machine Learning models of gene-metabolite relationships to discover cell types most relevant to each metabolite. Integrative analysis of disease severity scores along with multi-omics data revealed a shift in specific immune cell subtypes in low versus high grade disease. We conclude that leveraging gene-metabolite covariation in a patient cohort can uncover underlying molecular mechanisms, overcoming the challenges posed by high dimensionality of multi-omics data.

The lack of clinically predictive potency assays for cell products significantly impedes translation of these therapies. Here, we describe a microfluidic on-chip 3D system for rapid evaluation of a subset of patient-derived bone marrow aspirate concentrate (BMAC) samples used in a phase 3 multicenter trial (NCT03818737) evaluating autologous cells for relieving knee osteoarthritis pain. BMAC clinical samples cultured in the on-chip 3D system exhibit elevated levels of immunomodulatory and trophic proteins compared to 2D culture. Using analyte information from in vitro assays and patient-matched clinical data, we build linear regression prediction models for clinical outcomes. We demonstrate improved clinical prediction by cross-validation accuracy for the on-chip 3D platform compared to 2D culture. Additionally, on-chip 3D assay metrics display higher correlative power with patient pain scores compared to the 2D assay. This study establishes a potency assay with improved prediction power to accelerate translation of cell therapies.

The discovery of molecular relationships from high-dimensional data is a major open problem in bioinformatics. Machine learning and feature attribution models have shown great promise in this context but lack causal interpretation. Here, we show that a popular feature attribution model, under certain assumptions, estimates an average of a causal quantity reflecting the direct influence of one variable on another. We leverage this insight to propose a precise definition of a gene regulatory relationship and implement a new tool, CIMLA (Counterfactual Inference by Machine Learning and Attribution Models), to identify differences in gene regulatory networks between biological conditions, a problem that has received great attention in recent years. Using extensive benchmarking on simulated data, we show that CIMLA is more robust to confounding variables and is more accurate than leading methods. Last, we use CIMLA to analyze a previously published single-cell RNA sequencing dataset from subjects with and without Alzheimer’s disease (AD), discovering several potential regulators of AD.

Spatial transcriptomics has enabled the study of mRNA distributions within cells, a key aspect of cellular function. However, there is a dearth of tools that can identify and interpret functionally relevant spatial patterns of subcellular transcript distribution. To address this, we present CellSP, a computational framework for identifying, visualizing, and characterizing consistent subcellular spatial patterns of mRNA. CellSP introduces the concept of “gene-cell modules”, which are gene sets with coordinated subcellular transcript distributions in many cells. It provides intuitive visualizations of the captured patterns and offers functional insights into each discovered module. We demonstrate that CellSP reliably identifies functionally significant modules across diverse tissues and technologies. We use the tool to discover subcellular spatial phenomena related to myelination, axonogenesis and synapse formation in the mouse brain. We find immune response-related modules that change between kidney cancer and healthy samples, and myelination-related modules specific to mouse models of Alzheimer’s Disease.

Spatial transcriptomics (ST) technologies have enabled new explorations into the spatial organization of tissues and their functional implications. However, one of the most fundamental analyses – comparative analysis of spatial gene expression across phenotypes – remains a formidable challenge. We introduce SpaceExpress, a novel statistical tool for detecting phenotype-associated changes in spatial expression patterns. SpaceExpress employs a neural network to embed multiple ST samples in a common latent space, enabling robust cross-sample comparisons despite structural and technical variations. It then uses spline regression to test differential spatial expression of genes between conditions, identifying specific regions of the tissue where expression patterns diverge and quantifying the magnitude of those differences, with rigorous false discovery control and handling of multiple replicates per condition. It includes visualization tools to help interpret spatial pattern differences. We demonstrate the tool’s effectiveness on synthetic and real ST datasets, revealing mechanistic insights into behavior- and development-related neurogenomic changes in honey bees and mice. Our work extends the highly influential paradigm of differential gene expression analysis to spatial omics.

Predicting how genetic variation affects phenotypic outcomes at the organismal, cellular, and molecular levels requires deciphering the cis-regulatory code, the sequence rules by which non-coding regions regulate genes. In this perspective, we discuss recent computational progress and challenges towards solving this fundamental problem. We describe how cis-regulatory elements are mapped and how their sequence rules can be learned and interpreted with sequence-to-function neural networks, with the goal of identifying genetic variants in human disease. We also discuss how studies of the 3D chromatin organization could help identifying long-range regulatory effects and how current methods for mapping gene regulatory networks could better describe biological processes. We point out current gaps in knowledge along with technical limitations and benchmarking challenges of computational methods. Finally, we discuss newly emerging technologies, such as spatial transcriptomics, and outline strategies for creating a more general model of the cis-regulatory code that is more broadly applicable across cell types and individuals.