Band-based similarity indices for gene expression classification and clustering

Area of Science:

• Multivariate Data Analysis
• Computational Statistics
• Bioinformatics

Background:

• Traditional depth measures for multivariate data are computationally infeasible in high dimensions.
• Modified Band Depth (MBD) is an exception, proving effective for high-dimensional gene expression data analysis.
• Assessing data point centrality is crucial for understanding multivariate data structures.

Purpose of the Study:

• To develop and evaluate novel band-based similarity indices for multivariate data.
• To assess the computational efficiency and performance of these indices in high-dimensional settings.
• To compare the effectiveness of band-based similarity with classical distance metrics.

Main Methods:

• Utilized Modified Band Depth (MBD) to define centrality and relationships within data bands.
• Constructed binary matrices and contingency tables from data bands to quantify pairwise (dis)similarity.
• Derived standard similarity indices from contingency tables for analysis.

Main Results:

• The proposed band-based similarity approach is computationally efficient and scalable to high dimensions.
• Several derived similarity indices demonstrated superior performance compared to classical distances, including Euclidean distance.
• The method showed strong results in classification and clustering tasks across simulated and real datasets.

Conclusions:

• Band-based similarity indices derived from Modified Band Depth offer a powerful alternative for high-dimensional data analysis.
• The technique provides a computationally efficient and effective tool for uncovering patterns in complex datasets.
• The framework is flexible and extensible to other similarity coefficients and applications beyond classification/clustering.