Assisted gene expression-based clustering with AWNCut.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene expression (GE) data is crucial for complex disease research, aiding in disease subtype identification and risk stratification.
• Traditional clustering of GE data faces challenges due to small sample sizes and data noise, leading to unsatisfactory results.
• Multidimensional profiling, integrating GE with regulator data (e.g., copy number alterations, microRNAs, methylation), is a growing trend to capture richer biological insights.

Purpose of the Study:

• To develop a novel assisted clustering method that leverages regulator information to enhance GE data clustering.
• To improve the accuracy and biological relevance of sample clustering for complex diseases.
• To address the issue of noisy and non-informative GE data by implementing a data-dependent weighting strategy.

Main Methods:

• Developed a novel assisted clustering algorithm integrating gene expression and regulator data.
• Implemented a data-dependent weighted strategy to prioritize informative genes and mitigate noise.
• Utilized the Normalized Cut (NCut) technique and a simulated annealing algorithm for effective realization.

Main Results:

• The proposed method significantly outperforms existing clustering approaches in simulations.
• Analysis of TCGA cutaneous melanoma and lung adenocarcinoma data yielded biologically meaningful findings distinct from alternative methods.
• The weighted strategy effectively discriminates informative gene expression data from noise.

Conclusions:

• The novel assisted clustering method effectively integrates regulator information for improved GE data analysis.
• This approach enhances the identification of disease subtypes and stratification by improving clustering accuracy.
• The method demonstrates robust performance and biological relevance in real-world cancer datasets.