ILoReg

ILoReg identifies cell populations from scRNA-seq data with high resolution, particularly for populations with subtle transcriptomic differences.

Key Features:

• Iterative Clustering Projection (ICP): A self-supervised learning method that uses logistic regression and clustering similarity comparison to iteratively cluster data and find cluster probabilities.
• Probabilistic feature extraction: Cluster probabilities from ICP are used for noise reduction prior to PCA, hierarchical clustering, and t-SNE.
• L1-regularization: ICP integrates feature selection with clustering to identify genes differentially expressed between cell populations.
• Consensus solution: Multiple ICP runs are combined into a single consensus solution for high-resolution investigation of cell populations.
• Differential expression analysis: Functions are provided to find gene markers for identified cell populations.
• Gene expression visualization: Functions for visualizing gene expression are included.

Scientific Applications:

• Immune cell population identification: ILoReg enables more pronounced segregation of immune cell populations compared with state-of-the-art methods.
• Pancreatic cell population identification: ILoReg enables more pronounced segregation of pancreatic cell populations compared with state-of-the-art methods.
• High-resolution cell population identification: The tool improves identification of cell populations with subtle transcriptomic differences from single-cell RNA-seq data.

Methodology:

ILoReg uses iterative clustering projection (ICP), a self-supervised learning method employing logistic regression and clustering similarity comparison with L1-regularization for feature selection, followed by noise reduction using cluster probabilities, PCA, hierarchical clustering, and t-SNE.