scAEspy

scAEspy applies autoencoders to single-cell RNA sequencing (scRNA-Seq) data to learn low-dimensional representations and integrate datasets across sequencing platforms for improved cell-type identification.

Key Features:

• Autoencoder architectures: Implements five autoencoder architectures, including two novel architectures developed for scAEspy.
• Loss functions: Supports a variety of loss functions to optimize representation learning for gene expression data.
• Data integration: Integrates datasets generated from different scRNA-Seq platforms to produce harmonized low-dimensional representations.
• Batch-effect compatibility: Can be coupled with existing batch-effect removal tools to enhance cross-platform integration.
• Benchmarking: Demonstrated a greater than 20% increase in Rand Index for cell cluster identification relative to principal component analysis (PCA) on five public datasets.
• Modularity: Provides a modular architecture that allows incorporation of additional autoencoders.

Scientific Applications:

• Cell-type identification: Produces low-dimensional representations that improve identification of known and potentially novel cell types from scRNA-Seq data.
• Cross-platform data integration: Harmonizes heterogeneous scRNA-Seq datasets to support large-scale and multi-platform studies.

Methodology:

Uses autoencoders to model non-linear relationships in gene expression data; implements five autoencoder architectures (including two novel ones) and multiple loss functions; can be coupled with batch-effect removal tools; benchmarking performed against PCA on five public datasets using the Rand Index for cell cluster identification.