Identification of Interpretable Clusters and Associated Signatures in Breast Cancer Single-Cell Data: A Topic...

Area of Science:

• Computational biology
• Machine learning
• Genomics

Background:

• Topic modeling is a machine learning technique for text analysis, also applied to biological data for clustering and interpretation.
• Existing methods can cluster biological data but may lack interpretability or optimal gene partitioning.

Purpose of the Study:

• To develop a novel topic modeling approach for simultaneous clustering of single cells and detection of gene signatures (topics).
• To apply this method to analyze transcriptional heterogeneity in breast cancer, distinguishing drug-sensitive and resistant subtypes.

Main Methods:

• Developed a new topic modeling framework for multi-omics single-cell data.
• Applied the model to patient-derived xenograft models of breast cancer with acquired therapy resistance.
• Integrated messenger RNAs (mRNAs) and long non-coding RNAs (lncRNAs) for enhanced cell classification.

Main Results:

• Identified protein-coding genes and long non-coding RNAs (lncRNAs) that define distinct cell clusters.
• Successfully distinguished drug-sensitive and drug-resistant breast cancer subtypes.
• Demonstrated superior performance over standard clustering methods in terms of partitioning and interpretability.

Conclusions:

• The proposed topic modeling approach provides an interpretable and accurate method for single-cell data analysis.
• Integrative analysis of multiple omics layers, including lncRNAs, improves cell classification accuracy in cancer research.