ConGRI

ConGRI infers gene regulatory interactions from high-throughput spatial gene expression data, using contrastive deep learning on in situ hybridization images to reveal regulatory relationships.

Key Features:

• Contrastive Learning Scheme: Employs a contrastive learning approach to distinguish regulatory interactions by comparing pairs of gene expression images.
• Siamese CNN Architecture: Uses a deep Siamese convolutional neural network to process paired inputs and learn similarity and difference patterns indicative of regulation.
• Feature Embedding Extraction: Automatically extracts high-level feature embeddings from spatial expression patterns observed in images.
• Artificial Neural Network Prediction: Applies an artificial neural network to analyze embeddings and predict the existence of gene regulatory interactions.
• Input Data: Operates on high-throughput spatial gene expression data, specifically in situ hybridization images.

Scientific Applications:

• Drosophila embryogenesis GRN inference: Applied to a Drosophila embryogenesis dataset to infer gene regulatory networks associated with developmental processes.
• Early eye development GRN detection: Identified regulatory interactions for early eye development with reported accuracy of 76.7%.
• Mesoderm development GRN detection: Identified regulatory interactions for mesoderm development with reported accuracy of 68.7%.
• Master regulator identification: Contributed to the identification of master regulators in Drosophila eye development.

Methodology:

ConGRI applies contrastive learning with a deep Siamese CNN to extract high-level embeddings from paired in situ hybridization images and uses an artificial neural network to predict gene regulatory interactions.