Solo

Solo identifies doublets in single-cell RNA sequencing (scRNA-seq) datasets using a semi-supervised deep learning approach to improve the accuracy of single-cell measurements.

Key Features:

• Semi-supervised deep learning: Combines unsupervised and supervised learning components to distinguish doublets from singlets in scRNA-seq data.
• Variational autoencoder (VAE) embedding: Employs a variational autoencoder to embed cells into a latent space for dimensionality reduction while preserving underlying gene expression structure.
• Supervised classifier (feed-forward neural network): Appends a feed-forward neural network to the VAE encoder to perform cell-level classification.
• Training with simulated doublets and observed single-cell data: Trains the classifier using simulated doublets alongside observed single-cell profiles to learn doublet signatures.
• Per-cell doublet prediction: Produces cell-level predictions to enable filtering of doublets from scRNA-seq datasets.

Scientific Applications:

• scRNA-seq preprocessing: Filter doublets from single-cell datasets prior to downstream analyses such as clustering and differential expression.
• Complementary doublet detection: Serve as a computational filter to refine results from experimental doublet detection methods.
• Improving data quality for interpretation: Reduce doublet-induced artifacts to enhance biological interpretation of cell states and heterogeneity.

Methodology:

Uses a variational autoencoder (VAE) for unsupervised embedding into a latent space, then appends a feed-forward neural network to the VAE encoder to form a supervised classifier trained on simulated doublets and observed single-cell data.