Noise Resistant Generalized Parametric Validity Index of Clustering for Gene Expression Data

Area of Science:

• Data Mining
• Bioinformatics
• Machine Learning

Background:

• Clustering validity indices are crucial for assessing clustering performance.
• Existing indices lack evaluation for noise-resistance, especially for microarray data.
• Determining optimal clustering in noisy datasets remains a challenge.

Purpose of the Study:

• To propose a novel Generalized Parametric Validity (GPV) index.
• To enhance noise-resistance in clustering validity assessment.
• To provide guidelines for selecting parameters in the GPV index.

Main Methods:

• Developed a Generalized Parametric Validity (GPV) index with tunable parameters α and β.
• Evaluated the GPV index's noise-resistance on simulated gene expression data with varying noise levels.
• Compared the GPV index with eight existing validity indices for cluster number determination.
• Tested the GPV index on three real gene expression datasets.

Main Results:

• The proposed GPV index exhibits significant noise-resistance ability.
• Parameter tuning in the GPV index allows for flexibility in handling noisy data.
• The GPV index demonstrated superior performance compared to existing indices in noisy conditions.
• Accurate judgements were provided by the GPV index on real-world gene expression data.

Conclusions:

• The GPV index is a robust tool for cluster validity assessment, particularly in the presence of noise.
• The GPV index offers a reliable guideline for determining the best clustering in noisy datasets.
• The developed index advances the field of clustering analysis for high-dimensional, noisy biological data.