Back to Models

scGNN

✨ Generative Models

Graph Neural Network for Single-Cell

RNA

Graph autoencoder that models cell-cell relationships through cell graphs for imputation and clustering

Publications

scGNN is a novel graph neural network framework for single-cell RNA-Seq analyses

Wang et al.2021
Complexity
★★★
complex
Interpretability
★★☆
medium
Architecture
Graph Autoencoder
Latent Dim
16
Used in LAIOR Framework

Cell-Cell Graph Modeling

scGNN constructs cell-cell graphs and uses Graph Neural Networks to aggregate information across similar cells for imputation and clustering

Main Idea

Learn meaningful cell representations by modeling both gene expression and cell-cell similarity relationships with reconstruction

Key Components

Graph Construction

Build cell graphs using expression similarity

Graph Neural Network Encoder

Aggregate neighbor information via message passing

Decoder

Reconstruct missing/dropout values

Clustering Module

Identify cell populations from learned embeddings

Mathematical Formulation

H^(l+1) = σ(D^(-1/2) A D^(-1/2) H^(l) W^(l)); X̂ = Decoder(H_final)

Loss Functions

Reconstruction Loss
MSE(X, X̂)
Clustering Loss
Cross-Entropy

Data Flow

Gene Expression → Cell Graph Construction → GNN Encoder → Embeddings → Decoder → Imputation + Clustering

Architecture Details

Architecture Type

Graph Convolutional Autoencoder (VAE Architecture)

Input/Output Types

single-cell → reconstruction

Key Layers

GraphConvolutionAggregationMLPDecoder

Frameworks

PyTorch

Tags

graphclusteringimputationvaerna