Biology subjects

Blei, D.

Publications and source records attributed to Blei, D..

4 recordsLinked to original sources

bioRxiv2024

Variational inference for microbiome survey data with application to global ocean data

Linking sequence-derived microbial taxa abundances to host (patho-)physiology or habitat characteristics in a reproducible and interpretable manner has remained a formidable challenge for the analysis of microbiome survey data. Here, we introduce a flexible probabilistic modeling framework, VI-MIDAS (Variational Inference for MIcrobiome survey DAta analysiS), that enables joint estimation of context-dependent drivers and broad patterns of associations of microbial taxon abundances from microbiome survey data. VI-MIDAS comprises mechanisms for direct coupling of taxon abundances with covariates and taxa-specific latent coupling which can incorporate spatio-temporal information and taxon-taxon interactions. We leverage mean-field variational inference for posterior VI-MIDAS model parameter estimation and illustrate model building and analysis using Tara Ocean Expedition survey data. Using VI-MIDAS latent embedding model and tools from network analysis, we show that marine microbial communities can be broadly categorized into five modules, including SAR11-, Nitrosopumilus-, and Alteromondales-dominated communities, each associated with specific environmental and spatiotemporal signatures. VI-MIDAS also finds evidence for largely positive taxon-taxon associations in SAR11 or Rhodospirillales clades, and negative associations with Alteromonadales and Flavobacteriales classes. Our results indicate that VI-MIDAS provides a powerful integrative statistical analysis framework for discovering broad patterns of associations between microbial taxa and context-specific covariate data from microbiome survey data.

bioinformatics↗

bioRxiv2023

Deep generative model deciphers derailed trajectories in acute myeloid leukemia

Biological insights often depend on comparing conditions such as disease and health, yet we lack effective computational tools for integrating single-cell genomics data across conditions or characterizing transitions from normal to deviant cell states. Here, we present Decipher, a deep generative model that characterizes derailed cell-state trajectories. Decipher jointly models and visualizes gene expression and cell state from normal and perturbed single-cell RNA-seq data, revealing shared and disrupted dynamics. We demonstrate its superior performance across diverse contexts, including in pancreatitis with oncogene mutation, acute myeloid leukemia, and gastric cancer.

bioinformatics↗

bioRxiv2022

Starfysh reveals heterogeneous spatial dynamics in the breast tumor microenvironment

Spatially-resolved gene expression profiling provides valuable insight into tissue organization and cell-cell crosstalk; however, spatial transcriptomics (ST) lacks single-cell resolution. Current ST analysis methods require single-cell RNA sequencing data as a reference for a rigorous interpretation of cell states and do not utilize associated histology images. Significant sample variation further complicates the integration of ST datasets, which is essential for identifying commonalities across tissues or altered cellular wiring in disease. Here, we present Starfysh, the first comprehensive computational toolbox for joint modeling of ST and histology data, dissection of refined cell states, and systematic integration of multiple ST datasets from complex tissues. Starfysh uses an auxiliary deep generative model that incorporates archetypal analysis and any known cell state markers to avoid the need for a single-cell-resolution reference in characterizing known or novel tissue-specific cell states. Additionally, Starfysh improves the characterization of spatial dynamics in complex tissues by leveraging histology images and enables the comparison of niches as spatial "hubs" across tissues. Integrative analysis of primary estrogen receptor-positive (ER+) breast cancer, triple-negative breast cancer (TNBC), and metaplastic breast cancer (MBC) tumors using Starfysh led to the identification of heterogeneous patient- and disease-specific hubs as well as a shared stromal hub with varying spatial orientation. Our results show the ability to delineate the spatial co-evolution of tumor and immune cell states and their crosstalk underlying intratumoral heterogeneity in TNBC and revealed metabolic reprogramming shaping immunosuppressive hubs in aggressive MBC. Starfysh is publicly available (https://github.com/azizilab/starfysh).

bioinformatics↗

bioRxiv2018

The Markov link method: a nonparametric approach to combine observations from multiple experiments

This paper studies measurement linkage. An example from cell biology helps explain the problem: imagine for a given cell we can either sequence the cells RNA or we can examine its morphology, but not both. Given a cells morphology, what do we expect to see in its RNA? Given a cells RNA, what do we expect in its morphology? More broadly, given a measurement of one type, can we predict measurements of the other type? This measurement linkage problem arises in many scientific and technological fields. To solve this problem, we develop a nonparametric approach we dub the \"Markov link method\" (MLM). The MLM makes a conditional independence assumption that holds in many multi-measurement contexts and provides a way to estimate the link, the conditional probability of one type of measurement given the other. We derive conditions under which the MLM estimator is consistent and we use simulated data to show that it provides accurate measures of uncertainty. We evaluate the MLM on real data generated by a pair of single-cell RNA sequencing techniques. The MLM characterizes the link between them and helps connect the two notions of cell type derived from each technique. Further, the MLM reveals that some aspects of the link cannot be determined from the available data, and suggests new experiments that would allow for better estimates.\n\nSignificance StatementNovel experimental techniques are developing quickly, and each technique gives new perspectives. Ideally we would build theories that account for many perspectives at once. This is not easy. One challenge is that many experiments use measurement techniques that alter or destroy the subject, making it impossible to measure the same subject with both techniques and difficult to combine data from different experiments. In this paper we develop the Markov Link Method, a new tool that overcomes this challenge.

bioinformatics↗