SoCube

SoCube performs doublet detection in single-cell RNA sequencing (scRNA-seq) data to identify multiplet artifacts that confound downstream analyses such as differential expression and trajectory inference.

Key Features:

• Doublet detection: Identifies doublets arising when two or more cells are captured as a single observation in scRNA-seq data.
• Python implementation: Distributed as a Python package for integration into computational workflows.
• 3D composite feature-embedding: Uses a novel 3D composite feature-embedding strategy to capture latent gene information for classification.
• Multikernel, multichannel CNN ensemble: Core architecture is a multikernel, multichannel convolutional neural network (CNN) ensemble that processes embedded features.
• Generalization-focused design: Incorporates model architectures and embedding strategies intended to improve generalization across datasets.
• Benchmark performance: Evaluated on benchmark datasets and downstream tasks, demonstrating improved detection performance in reported evaluations.

Scientific Applications:

• Differential expression analysis: Removal of doublets to reduce false signals in differentially expressed gene analysis.
• Cell trajectory inference: Identification and removal of doublets to improve accuracy of cell trajectory and lineage reconstruction.
• scRNA-seq data quality control: Enhances data quality and reliability by detecting and flagging multiplet artifacts in scRNA-seq studies.

Methodology:

Implemented in Python; employs a deep learning algorithm using a 3D composite feature-embedding strategy and a multikernel, multichannel convolutional neural network (CNN) ensemble to capture latent gene information for doublet classification.