Wilson, C. M. | Search bioRxiv

bioRxiv2021

Quantification and visualization of the tumor microenvironment heterogeneity from spatial transcriptomic experiments

SummarySpatially-resolved transcriptomics promises to increase our understanding of the tumor microenvironment and improve cancer prognosis and therapies. Nonetheless, analytical methods to explore associations between the spatial heterogeneity of the tumor and clinical data are not available. Hence, we have developed spatialGE, a software that provides visualizations and quantification of the tumor microenvironment heterogeneity through gene expression surfaces, spatial heterogeneity statistics (SThet) that can be compared against clinical information, spot-level cell deconvolution, and spatially-informed clustering (STclust), all using a new data object to store data and resulting analyses simultaneously. Availability and implementationThe R package and tutorial/vignette are available at https://github.com/FridleyLab/spatialGE. A script to reproduce the analyses in this manuscript is available in Supplementary information. ContactFridley.Lab@Moffitt.org or Brooke.Fridley@Moffitt.org Supplementary informationAvailable at Bioinformatics online. O_FIG O_LINKSMALLFIG WIDTH=198 HEIGHT=200 SRC="FIGDIR/small/454023v2_figG1.gif" ALT="Figure 1"> View larger version (73K): org.highwire.dtl.DTLVardef@191cbe7org.highwire.dtl.DTLVardef@e1b6c7org.highwire.dtl.DTLVardef@a6f68forg.highwire.dtl.DTLVardef@1853601_HPS_FORMAT_FIGEXP M_FIG O_FLOATNOGraphical abstractC_FLOATNO Overview of spatialGE features. A. The STList data object from spatialGE can be creared from several sources, including comma- or tab-separated files containing gene counts and spatial coordinates. The object can also be created directly from Visium outputs, Seurat objects, or GeoMx outputs. B. Users can optionally provide a metadata file, containing information associated with each sample (one row per sample, or per ROI if GeoMx data). C. Methods for quality control of data are provided by spatialGE, including visualizations of counts and genes per spot, as well as filtering of spots or genes within user-determined thresholds. D. A novel method (STclust) performs spatially informed clustering of spots and tissue domain identification. E. spatialGE provides different types of data visualization, including gene expression at each spot ("quilt plots"), as well as adaptation of spatial interpolation ("kriging") to spatial transcriptomics data (transcriptomic surface). F. spatialGE also leverages spatial statistics (Morans I, Gearys C, Getis-Ord Gi) to quantitatively describe heterogeneity within the tumor microenvironment and to explore associations between spatial heterogeneity and clinical oucomes. G. Gene expression deconvolution can also be applied to each spot to detect immune cell types (xCell) and classification of spots as tumor or stroma (ESTIMATE). C_FIG

bioinformatics↗

bioRxiv2020

Wide and Deep Learning for Automatic Cell Type Identification

Cell type classification is an important problem in cancer research, especially with the advent of single cell technologies. Correctly identifying cells within the tumor microenvironment can provide oncologists with a snapshot of how a patients immune system is reacting to the tumor. Wide deep learning (WDL) is an approach to construct a cell-classification prediction model that can learn patterns within high-dimensional data (deep) and ensure that biologically relevant features (wide) remain in the final model. In this paper, we demonstrate that the use of regularization can prevent overfitting and adding a wide component to a neural network can result in a model with better predictive performance. In particular, we observed that a combination of dropout and{ell} 2 regularization can lead to a validation loss function that does not depend on the number of training iterations and does not experience a significant decrease in prediction accuracy compared to models with{ell} 1, dropout, or no regularization. Additionally, we show WDL can have superior classification accuracy when the training and testing of a model is completed data on that arise from the same cancer type, but from different platforms. More specifically, WDL compared to traditional deep learning models can substantially increase the overall cell type prediction accuracy (41 to 90%) and T-cell sub-types (CD4: 0 to 76%, and CD8: 61 to 96%) when the models were trained using melanoma data obtained from the 10X platform and tested on basal cell carcinoma data obtained using SMART-seq.

genomics↗