Cancer DNA microarray analysis considering multi-subclass with graph-based clustering method.

Area of Science:

• Genomics
• Bioinformatics
• Cancer Research

Background:

• Cancer development involves genetic factors (cell cycle, adhesion, transcription) and environmental influences (age, sex, lifestyle).
• The heterogeneity of cancer necessitates distinct patient groupings for accurate analysis.
• Identifying specific contributing factors to cancer onset is challenging due to multifactorial influences.

Purpose of the Study:

• To investigate the utility of graph-based clustering prior to classification analysis of DNA microarray datasets.
• To construct accurate multi-class classification models for various cancer types.
• To identify genes specific to distinct patient clusters.

Main Methods:

• Applied graph-based clustering to DNA microarray datasets.
• Utilized k-nearest neighbor for multi-class classification model construction.
• Employed One vs. Others classification to find cluster-specific genes.

Main Results:

• Achieved classification accuracies exceeding 80% for leukemia, breast, prostate, and colon cancer datasets.
• Successfully identified specific genes within a control group cluster in the breast cancer dataset.
• Demonstrated the effectiveness of pre-classification sample clustering.

Conclusions:

• Sample clustering is a crucial step before classification model construction in cancer genomics.
• This approach enhances the accuracy of cancer subtyping and gene discovery.
• The methodology holds promise for personalized medicine and targeted cancer therapies.