An enhanced deterministic K-Means clustering algorithm for cancer subtype prediction from gene expression data.

Area of Science:

• Biomedical informatics
• Machine learning
• Computational biology

Background:

• Non-deterministic clustering algorithms like K-Means, due to random initial centroid selection, yield inconsistent results.
• This inconsistency hinders the application of such algorithms in critical areas like cancer subtype prediction using gene expression data.
• Comparing results from non-deterministic algorithms with other methods is challenging.

Purpose of the Study:

• To address the limitations of non-deterministic clustering algorithms in biomedical applications.
• To develop a more stable and accurate clustering method for cancer subtype prediction.
• To improve the reliability and comparability of clustering results in gene expression analysis.

Main Methods:

• A novel, density-based K-Means clustering algorithm was developed.
• The algorithm employs a systematic method for selecting initial centroids from dense and well-separated regions in feature space.
• The proposed algorithm was compared against eleven established single clustering algorithms and one ensemble clustering algorithm.

Main Results:

• The proposed density-based K-Means algorithm demonstrated superior overall performance across ten cancer gene expression datasets.
• It consistently outperformed widely used single clustering algorithms and a prominent ensemble clustering algorithm.
• The algorithm provides stable and reproducible clustering results.

Conclusions:

• There is a significant need for reliable and user-friendly machine learning tools in cancer subtype prediction.
• The proposed algorithm is simple, easy to use, and provides stable predictions.
• It offers improved accuracy in predicting cancer subtypes from gene expression data, addressing a critical need in the biomedical domain.