CaFew

CaFew improves clustering of single-cell RNA sequencing (scRNA-seq) data by applying cluster-aware feature weighting to select genes that enhance cluster separation.

Key Features:

• Cluster-aware feature weighting: Assigns cluster-specific weights to genes to reflect each gene's contribution to separating clusters.
• Optimization of clustering objective: Optimizes a clustering objective function to derive the feature weight matrix.
• Feature weight matrix: Produces a feature weight matrix used to rank and select informative genes.
• Gene selection criteria: Selects genes with high weight in at least one cluster or with high weight variance across clusters.
• Integration with existing methods: Can be combined with clustering methods such as SC3, yielding improved clustering accuracy and reported state-of-the-art results.
• Visualization support: Highlights relevant gene features to aid visualization of scRNA-seq data.
• Empirical validation: Demonstrated performance improvements on eight real scRNA-seq datasets.

Scientific Applications:

• Improved clustering: Enhancing clustering accuracy for scRNA-seq datasets by selecting informative genes.
• Identification of distinguishing genes: Identifying genes that contribute to cluster separation via weight-based ranking.
• Data visualization: Supporting visualization of single-cell transcriptomic structure by highlighting relevant features.
• Workflow augmentation: Augmenting existing clustering workflows such as SC3 to achieve better cluster resolution.
• Cell heterogeneity exploration: Facilitating exploration of cellular heterogeneity in single-cell studies.

Methodology:

Applies an optimization strategy that optimizes a clustering objective function to derive a feature weight matrix and selects genes with high weights in at least one cluster or with high weight variance across clusters; evaluated on eight real scRNA-seq datasets and combined with SC3 in experiments.