A composite model for subgroup identification and prediction via bicluster analysis

Area of Science:

• Bioinformatics
• Computational Biology
• Data Science

Background:

• Analyzing large biomedical datasets presents challenges in subgroup identification, relationship characterization, and predictive modeling.
• Subgroups can represent diverse biological entities like pathogen serotypes, tumor subtypes, or patient genetic profiles influencing treatment response.

Purpose of the Study:

• To develop a composite model for subgroup identification and prediction using biclustering techniques.
• To create a classification model for assigning samples to distinct subgroups based on identified attributes.

Main Methods:

• Employs biclustering to identify biclusters within the data.
• Builds subgroup-specific binary classifiers for each bicluster.
• Constructs a composite model integrating these classifiers for sample classification into disjoint subgroups.

Main Results:

• Achieved 97.4% accuracy on a synthetic dataset with four subgroups.
• Demonstrated 83.7% accuracy in distinguishing lung cancer subtypes (adenocarcinoma vs. squamous carcinoma).
• Identified 5 serotypes and several subtypes in a pathogen dataset with approximately 94% accuracy.

Conclusions:

• Presents a novel biclustering-based classification model for unlabeled biomedical data.
• Combines unsupervised biclustering with supervised classification for subgroup analysis.
• Facilitates identification of unknown species or novel biomarkers for targeted therapies.