A multi-class predictor based on a probabilistic model: application to gene expression profiling-based diagnosis of...

Area of Science:

• Oncology
• Molecular Biology
• Bioinformatics

Background:

• Microscopic diagnosis is crucial but insufficient for certain cancer diagnostics, particularly differentiating malignant from benign thyroid tissues.
• Gene expression profiling is emerging as a valuable supplementary diagnostic tool for complex cases.

Purpose of the Study:

• To investigate the feasibility of gene expression profiling for molecular diagnosis of thyroid tissues.
• To develop and evaluate a novel multi-class predictor for accurate differential diagnosis of thyroid conditions.

Main Methods:

• Gene expression profiling was performed on four thyroid tissue types using adaptor-tagged competitive PCR.
• A novel multi-class predictor with probabilistic outputs was developed and trained on 119 samples.
• Predictor performance was rigorously evaluated using leave-one-out cross-validation and an independent test set of 49 samples.

Main Results:

• The multi-class predictor, utilizing extensive combinations of binary classifiers, demonstrated superior prediction accuracy compared to classical methods.
• The predictor achieved a test prediction accuracy of 85.7% on an independent dataset.
• Probabilistic outputs allowed for detailed sample information and visualization in low-dimensional classification spaces.

Conclusions:

• Molecular diagnosis of thyroid tissues via gene expression profiling is feasible and shows potential as an automated diagnostic aid.
• A multi-class predictor with exhaustive binary classifier combinations surpasses traditional methods and multi-class SVM in accuracy.
• The developed approach offers enhanced multi-class classification capabilities, beneficial for cancer diagnostics.