Biology subjects

Babadi, M.

Publications and source records attributed to Babadi, M..

16 recordsLinked to original sources

bioRxiv2026

A supervised ontology-aware cell annotation method forsingle-cell transcriptomic data

Many single-cell RNA-seq annotation methods ignore the hierarchical nature of cell type classification. We present a probability propagation strategy that enforces ontological consistency and improves performance when applied to existing models without retraining. Combined with a lightweight logistic regression model trained on 42 million human cells, this yields SOCAM, a fast and interpretable classifier. We also introduce a hop-based F1 score for ontology-aware evaluation. Code and models are available open source.

bioinformatics↗

bioRxiv2025

multiVIB: A unified probabilistic contrastive learning framework for atlas-scale integration of single-cell multi-omics data

Comprehensive brain cell atlases are essential for understanding neural function and advancing translational research. As single-cell technologies proliferate across platforms, species, and modalities, atlas construction increasingly depends on integration frameworks that are capable of aligning heterogeneous datasets without obscuring biological variations. However, existing methods are typically limited to narrow use cases, often requiring ad hoc workflows that may introduce artifacts. Here, we introduce multiVIB, a unified probabilistic contrastive learning framework for diverse scenarios of data integration. We demonstrate that multiVIB achieves state-of-the-art performance while minimizing spurious alignments. Applied to atlas-scale datasets generated by the BRAIN Initiative, multiVIB supports robust and scalable integration across data modalities and preserves species-specific variation in cross-species analyses. These results position multiVIB as a scalable, biologically faithful framework for the construction of next-generation brain cell atlases.

bioinformatics↗

bioRxiv2025

Accelerating scRNA-seq analysis: automated cell type annotation using representation learning and vector search

Cell type annotation in single-cell RNA sequencing (scRNA-seq) experiments is the fundamental step of assigning cell types to individual cells or clusters of cells based on their gene expression profiles. This process is crucial for developing biological insights from scRNA-seq experiments. We present a service that automates cell type annotation for 10x Genomics single-cell gene expression samples, enabling researchers to rapidly and accurately categorize cells within a sample. This service operates on the basis of reverse search: it compares each cells gene expression profile against the Chan Zuckerberg CELL by GENE (CZ CELLxGENE) Census, a comprehensive repository of published scRNA-seq datasets enriched with community-annotated cell types, and yields cell type annotations through summarizing the labels associated with similar cells. The annotation algorithm employed in this service avoids reliance on predefined marker genes or tissue-specific references, providing both fine-grained and coarse annotations. These initial annotations can be further refined by investigators to suit their specific research needs.

bioinformatics↗

bioRxiv2025

STPath: A Generative Foundation Model for Integrating Spatial Transcriptomics and Whole Slide Images

Spatial transcriptomics (ST) has shown remarkable promise in pathology applications, shedding light on the spatial organization of gene expression and its relationship to the tumor microenvironment. However, its clinical adoption remains constrained due to the limited scalability of current sequencing technologies. While recent methods attempt to infer ST from whole slide images (WSIs) using pretrained image encoders, they remain restricted by limited gene coverage, organ-specific training, and the need for dataset-specific fine-tuning. In light of this, we introduce STPath, a generative foundation model pretrained on a large-scale collection of WSIs paired with ST profiles. This extensive pretraining enables STPath to directly predict gene expression across 38,984 genes and 17 organs without requiring downstream fine-tuning. STPath integrates multiple data modalities, including histological images, gene expressions, organ type, and sequencing technology information, within a novel geometry-aware Transformer architecture. Unlike previous methods that directly map WSIs to gene expression, STPath is trained using a masked gene expression prediction objective guided by tailored noise schedules, effectively balancing between capturing gene-gene dependencies and performing high-quality predictions. We evaluate STPath across 6 tasks spanning 23 datasets and 14 biomarkers, including gene expression prediction, spot imputation, spatial clustering, biomarker prediction, gene mutation prediction, and survival prediction. These results demonstrate STPaths strong ability to infer spatially resolved gene expression and reveal crucial pathological structures within tissue samples, underscoring its promise for scalable ST-based pathology applications.

pathology↗

bioRxiv2024

TissueMosaic enables cross-sample differential analysis of spatial transcriptomics datasets through self-supervised representation learning

Spatial transcriptomics allows for the measurement of gene expression within native tissue context, thereby improving our understanding of how cell states are modulated by their microenvironment. Despite technological advancements, computational methods to link cell states with their microenvironment and perform comparative analysis across different samples and conditions are still underdeveloped. To address this, we introduce TissueMosaic (Tissue MOtif-based SpAtial Inference across Conditions), a self-supervised convolutional neural network designed to discover and represent tissue architectural motifs from multi-sample spatial transcriptomic datasets (https://github.com/broadinstitute/TissueMosaic). TissueMosaic effectively maps structurally similar tissue motifs close together in a learned latent space. TissueMosaic further links these motifs to gene expression, enabling the study of how changes in tissue structure impact function. TissueMosaic increases the signal-to-noise ratio of differential expression analysis through a motif enrichment strategy, resulting in more reliable detection of genes that covary with tissue structure. Here, we demonstrate TissueMosaic on high resolution spatial transcriptomics datasets across tissues, learning representations that outperform neighborhood cell-type composition baselines and existing methods on downstream tasks. We highlight genes and pathways in these tissues that are associated with changes in tissue structure across external conditions. These findings underscore the potential of self-supervised learning to significantly advance spatial transcriptomics research.

bioinformatics↗

bioRxiv2024

Robust self-supervised denoising of voltage imaging data using CellMincer

Voltage imaging enables high-throughput investigation of neuronal activity, yet its utility is often constrained by a low signal-to-noise ratio (SNR). Conventional denoising algorithms, such as those based on matrix factorization, impose limiting assumptions about the noise process and the spatiotemporal structure of the signal. While deep learning based denoising techniques offer greater adaptability, existing approaches fail to fully exploit the fast temporal dynamics and unique short- and long-range dependencies within voltage imaging datasets. Here, we introduce CellMincer, a novel self-supervised deep learning method designed specifically for denoising voltage imaging datasets. CellMincer operates on the principle of masking and predicting sparse sets of pixels across short temporal windows and conditions the denoiser on precomputed spatiotemporal auto-correlations to effectively model long-range dependencies without the need for large temporal denoising contexts. We develop and utilize a physics-based simulation framework to generate realistic datasets for rigorous hyperparameter optimization and ablation studies, highlighting the key role of conditioning the denoiser on precomputed spatiotemporal auto-correlations to achieve 3-fold further reduction in noise. Comprehensive benchmarking on both simulated and real voltage imaging datasets, including those with paired patch-clamp electrophysiology (EP) as ground truth, demonstrates CellMincers state-of-the-art performance. It achieves substantial noise reduction across the entire frequency spectrum, enhanced detection of subthreshold events, and superior cross-correlation with ground-truth EP recordings. Finally, we demonstrate how CellMincers addition to a typical voltage imaging data analysis workflow improves neuronal segmentation, peak detection, and ultimately leads to significantly enhanced separation of functional phenotypes.

neuroscience↗

bioRxiv2024

Modeling interpretable correspondence between cell state and perturbation response with CellCap

Single-cell transcriptomics, in conjunction with genetic and compound perturbations, offers a robust approach for exploring cellular behaviors in diverse contexts. Such experiments allow un-covering cell-state-specific responses to perturbations, a crucial aspect in unraveling the intricate molecular mechanisms governing cellular behavior and potentially discovering novel regulatory pathways and therapeutic targets. However, prevailing computational methods predominantly focus on predicting average cellular responses, disregarding the inherent response heterogeneity associated with cell state diversity. In this study, we present CellCap, a deep generative model designed for the end-to-end analysis of single-cell perturbation experiments. CellCap employs sparse dictionary learning in a latent space to deconstruct cell-state-specific perturbation responses into a set of transcriptional response programs. These programs are then utilized by each perturbation condition and each cell at varying degrees. The incorporation of specific model design choices, such as dot-product cross-attention between cell states and response programs, along with a linearly-decoded latent space, underlay the interpretation power of CellCap. We evaluate CellCaps model interpretability through multiple simulated scenarios and apply it to two real single-cell perturbation datasets. These datasets feature either heterogeneous cellular populations or a complex experimental setup. Our results demonstrate that CellCap successfully uncovers the relationship between cell state and perturbation response, unveiling novel insights overlooked in previous analyses. The models interpretability, coupled with its effectiveness in capturing heterogeneous responses, positions CellCap as a valuable tool for advancing our understanding of cellular behaviors in the context of perturbation experiments.

bioinformatics↗

bioRxiv2023

Capturing cell heterogeneity in representations of cell populations for image-based profiling using contrastive learning

Image-based cell profiling is a powerful tool that compares perturbed cell populations by measuring thousands of single-cell features and summarizing them into profiles. Typically a sample is represented by averaging across cells, but this fails to capture the heterogeneity within cell populations. We introduce CytoSummaryNet: a Deep Sets-based approach that improves mechanism of action prediction by 30-68% in mean average precision compared to average profiling on a public dataset. CytoSummaryNet uses self-supervised contrastive learning in a multiple-instance learning framework, providing an easier-to-apply method for aggregating single-cell feature data than previously published strategies. Interpretability analysis suggests that the model achieves this improvement by downweighting small mitotic cells or those with debris and prioritizing large uncrowded cells. The approach requires only perturbation labels for training, which are readily available in all cell profiling datasets. CytoSummaryNet offers a straightforward post-processing step for single-cell profiles that can significantly boost retrieval performance on image-based profiling datasets.

bioinformatics↗

bioRxiv2023

Transcriptional profile of the rat cardiovascular system at single cell resolution

BackgroundDespite the critical role of the cardiovascular system, our understanding of its cellular and transcriptional diversity remains limited. We therefore sought to characterize the cellular composition, phenotypes, molecular pathways, and communication networks between cell types at the tissue and sub-tissue level across the cardiovascular system of the healthy Wistar rat, an important model in preclinical cardiovascular research. We obtained high quality tissue samples under controlled conditions that reveal a level of cellular detail so far inaccessible in human studies. Methods and ResultsWe performed single nucleus RNA-sequencing in 78 samples in 10 distinct regions including the four chambers of the heart, ventricular septum, sinoatrial node, atrioventricular node, aorta, pulmonary artery, and pulmonary veins (PV), which produced an aggregate map of 505,835 nuclei. We identified 26 distinct cell types and additional subtypes, including a number of rare cell types such as PV cardiomyocytes and non-myelinating Schwann cells (NMSCs), and unique groups of vascular smooth muscle cells (VSMCs), endothelial cells (ECs) and fibroblasts (FBs), which gave rise to a detailed cell type distribution across tissues. We demonstrated differences in the cellular composition across different cardiac regions and tissue-specific differences in transcription for each cell type, highlighting the molecular diversity and complex tissue architecture of the cardiovascular system. Specifically, we observed great transcriptional heterogeneities among ECs and FBs. Importantly, several cell subtypes had a unique regional localization such as a subtype of VSMCs enriched in the large vasculature. We found the cellular makeup of PV tissue is closer to heart tissue than to the large arteries. We further explored the ligand-receptor repertoire across cell clusters and tissues, and observed tissue-enriched cellular communication networks, including heightened Nppa - Npr1/2/3 signaling in the sinoatrial node. ConclusionsThrough a large single nucleus sequencing effort encompassing over 500,000 nuclei, we broadened our understanding of cellular transcription in the healthy cardiovascular system. The existence of tissue-restricted cellular phenotypes suggests regional regulation of cardiovascular physiology. The overall conservation in gene expression and molecular pathways across rat and human cell types, together with our detailed transcriptional characterization of each cell type, offers the potential to identify novel therapeutic targets and improve preclinical models of cardiovascular disease.

cell biology↗

bioRxiv2022

A single-nucleus and spatial transcriptomic atlas of the COVID-19 liver reveals topological, functional, and regenerative organ disruption in patients

The molecular underpinnings of organ dysfunction in acute COVID-19 and its potential long-term sequelae are under intense investigation. To shed light on these in the context of liver function, we performed single-nucleus RNA-seq and spatial transcriptomic profiling of livers from 17 COVID-19 decedents. We identified hepatocytes positive for SARS-CoV-2 RNA with an expression phenotype resembling infected lung epithelial cells. Integrated analysis and comparisons with healthy controls revealed extensive changes in the cellular composition and expression states in COVID-19 liver, reflecting hepatocellular injury, ductular reaction, pathologic vascular expansion, and fibrogenesis. We also observed Kupffer cell proliferation and erythrocyte progenitors for the first time in a human liver single-cell atlas, resembling similar responses in liver injury in mice and in sepsis, respectively. Despite the absence of a clinical acute liver injury phenotype, endothelial cell composition was dramatically impacted in COVID-19, concomitantly with extensive alterations and profibrogenic activation of reactive cholangiocytes and mesenchymal cells. Our atlas provides novel insights into liver physiology and pathology in COVID-19 and forms a foundational resource for its investigation and understanding.

genomics↗

bioRxiv2022

GATK-gCNV: A Rare Copy Number Variant Discovery Algorithm and Its Application to Exome Sequencing in the UK Biobank

Copy number variants (CNVs) are major contributors to genetic diversity and disease. To date, exome sequencing (ES) has been generated for millions of individuals in international biobanks, human disease studies, and clinical diagnostic screening. While standardized methods exist for detecting short variants (single nucleotide and insertion/deletion variants) using tools such as the Genome Analysis ToolKit (GATK), technical challenges have confounded similarly uniform large-scale CNV analyses from ES data. Given the profound impact of rare and de novo coding CNVs on genome organization and human disease, the lack of widely-adopted and robustly benchmarked rare CNV discovery tools has presented a barrier to routine exome-wide assessment of this critical class of variation. Here, we introduce GATK-gCNV, a flexible algorithm to discover rare CNVs from genome sequencing read-depth information, which we distribute as an open-source tool packaged in GATK. GATK-gCNV uses a probabilistic model and inference framework that accounts for technical biases while simultaneously predicting CNVs, which enables self-consistency between technical read-depth normalization and variant calling. We benchmarked GATK-gCNV in 7,962 exomes from individuals in quartet families with matched genome sequencing and microarray data. These analyses demonstrated 97% recall of rare ([≤]1% site frequency) coding CNVs detected by microarrays and 95% recall of rare coding CNVs discovered by genome sequencing at a resolution of more than two exons. We applied GATK-gCNV to generate a reference catalog of rare coding CNVs in 197,306 individuals with ES from the UK Biobank. We observed strong correlations between CNV rates per gene and measures of mutational constraint, as well as rare CNV associations with multiple traits. In summary, GATK-gCNV is a tunable approach for sensitive and specific CNV discovery in ES, which can easily be applied across trait association and clinical screening.

genomics↗

bioRxiv2022

Learning representations for image-based profiling of perturbations

Measuring the phenotypic effect of treatments on cells through imaging assays is an efficient and powerful way of studying cell biology, and requires computational methods for transforming images into quantitative data that highlight phenotypic outcomes. Here, we present an optimized strategy for learning representations of treatment effects from high-throughput imaging data, which follows a causal framework for interpreting results and guiding performance improvements. We use weakly supervised learning (WSL) for modeling associations between images and treatments, and show that it encodes both confounding factors and phenotypic features in the learned representation. To facilitate their separation, we constructed a large training dataset with Cell Painting images from five different studies to maximize experimental diversity, following insights from our causal analysis. Training a WSL model with this dataset successfully improves downstream performance, and produces a reusable convolutional network for image-based profiling, which we call Cell Painting CNN-1. We conducted a comprehensive evaluation of our strategy on three publicly available Cell Painting datasets, discovering that representations obtained by the Cell Painting CNN-1 can improve performance in downstream analysis for biological matching up to 30% with respect to classical features, while also being more computationally efficient.

bioinformatics↗

bioRxiv2022

Uncovering features of synapses in primary visual cortex through contrastive representation learning

3D EM connectomics image volumes are now surpassing sizes of 1 mm3, and are therefore beginning to contain multiple meaningful spatial scales of brain circuitry simultaneously. However, the sheer density of information in such datasets makes the development of unbiased, scalable machine learning techniques a necessity for extracting novel insights without extremely time-consuming, intensive labor. In this paper, we present SynapseCLR, a self-supervised contrastive representation learning method for 3D electron microscopy (EM) data, and use the method to extract feature representations of synapses from a 3D EM dataset from mouse visual cortex. We show that our representations separate synapses according to both their overall physical appearance and structural annotations of known functional importance. We further demonstrate the utility of our methodology for several valuable downstream tasks for the growing field of 3D EM connectomics. These include one-shot identification of defective synapse segmentations, dataset-wide similarity-based querying, and accurate imputation of annotations for unlabeled synapses, using only manual annotation of 0.2% of synapses in the dataset. In particular, we show that excitatory vs. inhibitory neuronal cell types can be assigned to individual synapses and highly truncated neurites with accuracy exceeding 99.8%, making this population accessible to connectomics analysis. Finally, we present a data-driven and unsupervised study of the manifold of synaptic structural variation, revealing its intrinsic axes of variation and showing that synapse structure is also strongly correlated with inhibitory neuronal subtypes.

neuroscience↗

bioRxiv2021

High-throughput RNA isoform sequencing using programmable cDNA concatenation

Alternative splicing is a core biological process that enables profound and essential diversification of gene function. Short-read RNA sequencing approaches fail to resolve RNA isoforms and therefore primarily enable gene expression measurements - an isoform unaware representation of the transcriptome. Conversely, full-length RNA sequencing using long-read technologies are able to capture complete transcript isoforms, but their utility is deeply constrained due to throughput limitations. Here, we introduce MAS-ISO-seq, a technique for programmably concatenating cDNAs into single molecules optimal for long-read sequencing, boosting the throughput >15 fold to nearly 40 million cDNA reads per run on the Sequel IIe sequencer. We validated unambiguous isoform assignment with MAS-ISO-seq using a synthetic RNA isoform library and applied this approach to single-cell RNA sequencing of tumor-infiltrating T cells. Results demonstrated a >30 fold boosted discovery of differentially spliced genes and robust cell clustering, as well as canonical PTPRC splicing patterns across T cell subpopulations and the concerted expression of the associated hnRNPLL splicing factor. Methods such as MAS-ISO-seq will drive discovery of novel isoforms and the transition from gene expression to transcript isoform expression analyses.

genetics↗

bioRxiv2020

Transcriptional and Cellular Diversity of the Human Heart

IntroductionThe human heart requires a complex ensemble of specialized cell types to perform its essential function. A greater knowledge of the intricate cellular milieu of the heart is critical to increase our understanding of cardiac homeostasis and pathology. As recent advances in low input RNA-sequencing have allowed definitions of cellular transcriptomes at single cell resolution at scale, here we have applied these approaches to assess the cellular and transcriptional diversity of the non-failing human heart. MethodsMicrofluidic encapsulation and barcoding was used to perform single nuclear RNA sequencing with samples from seven human donors, selected for their absence of overt cardiac disease. Individual nuclear transcriptomes were then clustered based upon transcriptional profiles of highly variable genes. These clusters were used as the basis for between-chamber and between-sex differential gene expression analyses and intersection with genetic and pharmacologic data ResultsWe sequenced the transcriptomes of 287,269 single cardiac nuclei, revealing a total of 9 major cell types and 20 subclusters of cell types within the human heart. Cellular subclasses include two distinct groups of resident macrophages, four endothelial subtypes, and two fibroblasts subsets. Comparisons of cellular transcriptomes by cardiac chamber or sex reveal diversity not only in cardiomyocyte transcriptional programs, but also in subtypes involved in extracellular matrix remodeling and vascularization. Using genetic association data, we identified strong enrichment for the role of cell subtypes in cardiac traits and diseases. Finally, intersection of our dataset with genes on cardiac clinical testing panels and the druggable genome reveals striking patterns of cellular specificity. ConclusionsUsing large-scale single nuclei RNA sequencing, we have defined the transcriptional and cellular diversity in the normal human heart. Our identification of discrete cell subtypes and differentially expressed genes within the heart will ultimately facilitate the development of new therapeutics for cardiovascular diseases.

genomics↗

bioRxiv2019

CellBender remove-background: a deep generative model for unsupervised removal of background noise from scRNA-seq datasets

Droplet-based single-cell assays, including scRNA-seq, snRNA-seq, and CITE-seq, produce a significant amount of background noise counts, the hallmark of which is non-zero counts in cell-free droplets and off-target gene expression in unexpected cell types. The presence of such systematic background noise is a potential source of batch effect and spurious differential gene expression. Here we develop a deep generative model for noise-contaminated data that is structured to reflect the phenomenology of background noise generation in droplet-based single-cell assays. The proposed model successfully distinguishes cell-containing from cell-free droplets without supervision, learns the profile of background noise, and retrieves a noise-free quantification in an end-to-end fashion. We present a scalable and robust implementation of our method as a module in the open-source software package CellBender. We show that CellBender operates close to the theoretically optimal denoising limit in simulated datasets, and present extensive evaluations using real datasets and experimental benchmarks drawn from different tissues, protocols, and modalities to show that CellBender significantly improves the agreement of droplet-based single-cell data with established gene expression patterns, and that the learned background noise profile provides evidence for degraded or uncaptured cell types.

bioinformatics↗