Model Architecture Analysis and Implementation of TENET for Cell-Cell Interaction Network Reconstruction Using Spatial Transcriptomics Data.

Abstract

Cellular communication relies on the intricate interplay of signaling molecules, which come together to form the cell-cell interaction (CCI) network that orchestrates tissue behavior. Researchers have shown that shallow neural networks can effectively reconstruct the CCI from the abundant molecular data captured in spatial transcriptomics (ST). However, in scenarios characterized by sparse connections and excessive noise within the CCI, shallow networks are often susceptible to inaccuracies, leading to suboptimal reconstruction outcomes. To achieve a more comprehensive and precise CCI reconstruction, we propose a novel method called triple-enhancement-based graph neural network (TENET). The TENET framework has been implemented and evaluated on both real and synthetic ST datasets. This protocol primarily introduces our network architecture and its implementation. Key features • Cell-cell reconstruction network using ST data. • To facilitate the implementation of a more holistic CCI, we incorporate diverse CCI modalities into consideration. • To further enrich the input information, the downstream gene regulatory network (GRN) is also incorporated as an input to the network. • The network architecture considers global and local cellular and genetic features rather than solely leveraging the graph neural network (GNN) to model such information.