Integrative Approaches for Spatial and Multi-Omics data

点击翻译按钮获取中文摘要

Genomic technologies combined with modern data analytics are powerful approaches to revealing new insights into various biological questions. Recently, those approaches have expanded to the spatial dimension where thousands of genes are measured at each point across a 2D tissue sample. However, few tools exist to integrate the analysis of large amounts of biological data with spatial information. Instead, the spatial analyses are left to hands-on annotation by an expert biologist. Thus, there is a gap in developing analysis tools for spatial multi-omics that learn biologically meaningful-latent variables and use those variables for inference and prediction. Spatial multi-omics consists of count matrices from multiple modalities. Some modalities have partially overlapping variables, a modality with count matrix and spatial information, while others have a natural grouping of variables. Thus, we can consider that spatial multi-omics is generated from an underlying latent variable process that captures co-occurrence, non-linear interaction, and natural grouping of variables within and across modalities. The overarching long-term goal of my research program is to 1) develop transformative statistical and machine learning methods that integrate complex, real-world data across multiple modalities, 2) advance knowledge in characterizing their performance. For example, my recent studies have shown that probabilistic latent Dirichlet allocation identifies functional synonyms variables represented in bacterial communities. The main objective of this Discovery grant is to develop integrative approaches to discover latent variables in spatial multi-omics and use those variables to build Bayesian hierarchical models for prediction and inference. The studies extend our current methods development into microbiome multi-omics, leading to latent variable and spatial features extraction. The short-term objectives of this research are to 1. Investigate probabilistic graphical models for spatial multi-omics, including semi-supervised methods, 2. Develop Bayesian hierarchical models for prediction and inference, 3. Investigate Bayesian sampling methods and model assessment for spatial multi-omics integrative approaches. We have spatial multi-omics data from my collaborators for the human brain and gut microbiome. Having two datasets will allow us to evaluate the performance of the integrative approaches using domain experts' knowledge, reference models, and simulation studies. The studies will lead to new analytical tools for uncovering biological heterogeneity and spatial features for prediction and inference in spatial multi-omics. The tools will be available as open-source libraries to facilitate their use. Our work is potentially transformative to multiple modalities of data, including space. Thus, this work will impact many disciplines such as biology, ecology, agriculture, finance.