Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HistoSPACE: Histology-Inspired Spatial Transcriptome Prediction And Characterization Engine

Shivam Kumar, Samrat Chatterjee

TL;DR

A model is developed that explores the diversity of histological images available with ST data to extract molecular insights from tissue images and allows us to link the predicted expression with disease pathology.

Abstract

Spatial transcriptomics (ST) enables the visualization of gene expression within the context of tissue morphology. This emerging discipline has the potential to serve as a foundation for developing tools to design precision medicines. However, due to the higher costs and expertise required for such experiments, its translation into a regular clinical practice might be challenging. Despite the implementation of modern deep learning to enhance information obtained from histological images using AI, efforts have been constrained by limitations in the diversity of information. In this paper, we developed a model, HistoSPACE that explore the diversity of histological images available with ST data to extract molecular insights from tissue image. Our proposed study built an image encoder derived from universal image autoencoder. This image encoder was connected to convolution blocks to built the final model. It was further fine tuned with the help of ST-Data. This model is notably lightweight in compared to traditional histological models. Our developed model demonstrates significant efficiency compared to contemporary algorithms, revealing a correlation of 0.56 in leave-one-out cross-validation. Finally, its robustness was validated through an independent dataset, showing a well matched preditction with predefined disease pathology.

HistoSPACE: Histology-Inspired Spatial Transcriptome Prediction And Characterization Engine

TL;DR

A model is developed that explores the diversity of histological images available with ST data to extract molecular insights from tissue images and allows us to link the predicted expression with disease pathology.

Abstract

Spatial transcriptomics (ST) enables the visualization of gene expression within the context of tissue morphology. This emerging discipline has the potential to serve as a foundation for developing tools to design precision medicines. However, due to the higher costs and expertise required for such experiments, its translation into a regular clinical practice might be challenging. Despite the implementation of modern deep learning to enhance information obtained from histological images using AI, efforts have been constrained by limitations in the diversity of information. In this paper, we developed a model, HistoSPACE that explore the diversity of histological images available with ST data to extract molecular insights from tissue image. Our proposed study built an image encoder derived from universal image autoencoder. This image encoder was connected to convolution blocks to built the final model. It was further fine tuned with the help of ST-Data. This model is notably lightweight in compared to traditional histological models. Our developed model demonstrates significant efficiency compared to contemporary algorithms, revealing a correlation of 0.56 in leave-one-out cross-validation. Finally, its robustness was validated through an independent dataset, showing a well matched preditction with predefined disease pathology.
Paper Structure (20 sections, 6 equations, 7 figures, 1 algorithm)

This paper contains 20 sections, 6 equations, 7 figures, 1 algorithm.

Table of Contents

  1. Introduction
  2. Literature review
  3. Autoencoders for Images
  4. Expression prediction
  5. Methodology
  6. Image Autoencoder
  7. Building expression prediction model
  8. Selecting the loss functions
  9. Performance evaluation using correlation
  10. Experiments and Evaluations
  11. Datasets
  12. ICIAR Dataset
  13. STNet Dataset
  14. HER2 Dataset
  15. Image standardization technique
  16. ...and 5 more sections

Figures (7)

  • Figure 1: Brain of HistoSPACE: This is a complete framework consisting of two major components, an image autoencoder and an expression prediction model. The upper portion shows the image encoder-decoder network. The lower left shows the components of the convolution block, which are convolution, batch normalization, Relu and maximum pooling layers. The lower right shows, where encoder is connected with convolution blocks to predict the spatial expression.
  • Figure 2: Effect of image normalization. (A) Raw image. (B) Color cast removed. (C) Stain normalized with the reference image. (D) Combine the effect of color cast removal and stain normalization.
  • Figure 3: Building the autoencoder model. (A-C) The learning curve of training and test data for the three models, basic to advanced. (D-F) The prediction images from the testing tile set for the basic to advanced models show better imaging features captured.
  • Figure 4: Using encoder to build the final model by extending over a few CNN layers. (A) Less complex model showing better learning curve. (B) A more complex model shows an inconsistent relation in training testing data.
  • Figure 5: Comparison of HistoSPACE with STNet with leaving one out validation scheme. (A) Growing learning curve of STNet till 50 epochs. (B) Learning saturation achieved by our model in 50 epochs. (C-D) Expression distribution of both the algorithms for non-cancer/cancer tiles in particular samples.
  • ...and 2 more figures