scGMAI

scGMAI performs clustering of single-cell RNA-Seq (scRNA-Seq) data to identify cell types by combining deep autoencoder-based reconstruction, FastICA dimensionality reduction, and Gaussian mixture model clustering to mitigate dropout effects and high-dimensional noise.

Key Features:

• Gaussian Mixture Model-Based Clustering: Applies a Gaussian mixture model to cluster cells by gene expression profiles for identification and classification of distinct cell types.
• Deep Autoencoder Networks: Uses deep autoencoder networks to reconstruct the gene expression matrix from raw scRNA-Seq data and mitigate dropout events.
• Fast Independent Component Analysis (FastICA): Applies FastICA to the reconstructed expression data for dimensionality reduction to address high dimensionality and improve clustering efficiency.
• Comparative Performance: Demonstrated superior clustering performance relative to Seurat across 17 public scRNA-Seq datasets, yielding more accurate cell-type identification.

Scientific Applications:

• Cell Type Identification: Accurately distinguishes between different cell populations within heterogeneous samples.
• Developmental Biology Studies: Analyzes gene expression changes during cellular differentiation and development processes.
• Cancer Research: Investigates tumor heterogeneity and identifies subpopulations of cancer cells with distinct molecular profiles.

Methodology:

Autoencoder networks are trained to reconstruct the scRNA-Seq gene expression matrix to address dropout events, FastICA is applied to the reconstructed data for dimensionality reduction, and clustering is performed using a Gaussian mixture model.