scSemiAE

scSemiAE performs semi-supervised dimensionality reduction of single-cell RNA sequencing (scRNA-seq) data to produce low-dimensional representations that preserve cell subpopulation structure and support downstream analyses.

Key Features:

• Semi-supervised learning: Leverages both labeled and unlabeled scRNA-seq data, using cell subpopulation labels to guide dimensionality reduction.
• Autoencoder-based model: Implements an autoencoder neural network to learn efficient low-dimensional codings from high-dimensional scRNA-seq input.
• Transfer learning capability: Transfers information from labeled datasets to improve dimensionality reduction on other datasets when labeled cells are limited.
• Preservation of biological signal: Produces refined low-dimensional representations intended to retain essential biological variation for downstream analyses.

Scientific Applications:

• Clustering and cell subpopulation identification: Enhances clustering algorithms' ability to identify distinct cell subpopulations from scRNA-seq data.
• Improved downstream analysis: Facilitates downstream tasks such as differential expression analysis, trajectory inference, and integration of multiple single-cell datasets.

Methodology:

scSemiAE uses an autoencoder-based semi-supervised dimensionality reduction method that integrates labeled and unlabeled scRNA-seq data and incorporates transfer learning to leverage labeled datasets for improving representations of other datasets.