Towards clustering of incomplete microarray data without the use of imputation

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Gene expression data analysis often involves clustering to identify co-regulated genes.
• Missing values in microarray data can significantly impact clustering accuracy and reliability.
• Conventional clustering methods rely on pre-processing imputation, which can introduce errors that persist throughout analysis.

Purpose of the Study:

• To develop a novel clustering method that iteratively improves missing value estimates.
• To overcome the limitations of pre-processing imputation in clustering incomplete gene expression data.
• To enhance the biological relevance of clustering results from datasets with missing values.

Main Methods:

• Clustering Incomplete data using Alternating Optimization (CIAO) was developed, avoiding a separate imputation step.
• CIAO employs an alternating optimization approach, refining missing value estimates using cluster centroids and available data within each iteration.
• The method was evaluated on two yeast datasets and compared against k-means with KNNimpute.

Main Results:

• CIAO demonstrated improved estimation of missing values by leveraging cluster information iteratively.
• Clustering results from CIAO showed significantly higher relevance to Saccharomyces Genome Database gene annotations compared to conventional methods.
• The study highlights the effectiveness of CIAO for analyzing incomplete gene expression datasets.

Conclusions:

• CIAO offers a robust alternative to imputation-based clustering for gene expression data.
• The iterative refinement of missing values in CIAO leads to more biologically meaningful clustering outcomes.
• This method holds significant potential for advancing the analysis of large-scale, incomplete biological datasets.