Efficient prediction of a spatial transcriptomics profile better characterizes breast cancer tissue sections without costly experimentation

Abstract

Spatial transcriptomics is an emerging technology requiring costly reagents and considerable skills, limiting the identification of transcriptional markers related to histology. Here, we show that predicted spatial gene-expressions in unmeasured regions and tissues can enhance biologists' histological interpretations. We developed the Deep learning model for Spatial gene Clusters and Expression, DeepSpaCE and confirmed its performance using the spatial-transcriptome profiles and immunohistochemistry images of consecutive human breast cancer tissue sections. For example, the predicted expression patterns of SPARC, an invasion marker, highlighted a small tumor-invasion region that is difficult to identify using raw data of spatial transcriptome alone because of a lack of measurements. We further developed semi-supervised DeepSpaCE using unlabeled histology images and increased the imputation accuracy of consecutive sections, enhancing applicability for a small sample size. Our method enables users to derive hidden histological characters via spatial transcriptome and gene annotations, leading to accelerated biological discoveries without additional experiments.

Deep topographic proteomics of a human brain tumour

“…This bulk property means that the data still suffers from missing values which can confound spatial pathway level analysis. Future approaches are bound to use systematic spatial proteomic analysis, possibly compromising spatial resolution but incorporating an element of machine learning to use orthogonal higher resolution omics and imaging data to infer protein abundance towards individual cell resolution, as can be done on spatial transcriptomics data [58][59][60][61] . In addition, with the detection of LCM-based and cell-type resolved deep proteomes, these data will be highly complementary to current imaging technologies and increase the understanding of spatially resolved biological and pathological processes at the molecular level 29,32,57 .…”

Detection of Abnormal Extraocular Muscles in Small Datasets of Computed Tomography Images Using a Three–dimensional Variational Autoencoder: A Pilot Study

“…Most DL studies involving small numbers of medical images used supervised, [19][20][21] or semi-supervised DL. [22][23][24] These approaches may be used because supervised DL algorithms are expected to identify features that distinguish among data in small datasets; they require de nite answers to focus on during the model training. Unsupervised DL also extracts and learns features from input data.…”

Prediction of cancer treatment response from histopathology images through imputed transcriptomics

“…Key advances are already underway, as whole slide images (WSI) of tissue stained with hematoxylin and eosin (H&E) have been used to computationally diagnose tumors [1][2][3], classify cancer types [3][4][5][6][7][8], distinguish tumors with low or high mutation burden [9], identify genetic mutations [2,[10][11][12][13][14][15][16][17], predict patient survival [18][19][20][21][22], detect DNA methylation patterns [23] and mitoses [24], and quantify tumor immune infiltration [25]. Moreover, the ability to infer gene expression from WSI has also been explored [26][27][28][29][30].…”

“…To overcome this critical challenge, we introduce here for the first time a generic methodology for generating WSI-based predictors of patients' response for a broad range of cancer types and therapies, which does not require matched WSI and response datasets for training. To accomplish this, we have taken a two-step approach: First, we developed DeepPT (Deep Pathology for Transcriptomics), a novel deep-learning framework for imputing (predicting) gene expression from H&E slides, which extends upon previous valuable work on this topic [37][38][39][40][41]. The DeepPT models are cancer type-specific and are built by analyzing matched WSI and expression data from the TCGA.…”