netAE

netAE performs semi-supervised dimensionality reduction of single-cell RNA sequencing (scRNA-seq) data to propagate gold-standard labels and preserve intrinsic similarity structure for improved cell labeling.

Key Features:

• Semi-Supervised Label Propagation: Leverages a small number of gold-standard cell labels to guide dimensionality reduction and extend labels across the dataset.
• Network-Enhanced Autoencoder Architecture: Employs an autoencoder enhanced by network-based methods to maintain and utilize the intrinsic similarity structure of the original high-dimensional scRNA-seq space.
• Preservation of Data Structure: Preserves similarity relationships from the original high-dimensional scRNA-seq data during dimensionality reduction.
• Improved Labeling Accuracy: Achieves higher cell labeling accuracy compared to traditional unsupervised clustering methods when labeled samples are limited.
• Benchmarking: Validated on three public scRNA-seq datasets, demonstrating superior performance over baseline dimensionality reduction methods.

Scientific Applications:

• Cell population identification and classification: Facilitates identification and classification of known and novel cell populations within single-cell RNA sequencing datasets.
• Semi-supervised cell labeling with limited annotations: Enables accurate cell labeling in settings where comprehensive annotation is impractical by leveraging scarce gold-standard labels.

Methodology:

Identifies a small set of gold-standard labels which are used to inform a network-enhanced autoencoder-based dimensionality reduction that propagates labels across the dataset.