cancerclass

cancerclass implements methods to develop and validate diagnostic, prognostic, and predictive classifiers for cancer using high-dimensional molecular data.

Key Features:

• Feature Selection: Identifies relevant biomarkers from high-dimensional molecular data prior to classifier construction.
• Nearest-Centroid Classification: Performs sample classification using nearest-centroid classifiers.
• Validation Methods: Supports Training and Test set validation, Leave-One-Out Cross-Validation, and multiple random validation protocols including an extended multiple random validation protocol.
• Continuous Prediction Scores: Calculates and visualizes continuous prediction scores to enable thresholding and clinical interpretation.
• Score-to-Probability Conversion: Converts prediction scores into probabilities to distinguish clear-cut versus borderline classification results.
• Clinical Thresholding: Provides mechanisms to balance sensitivity and specificity and to define cutoff values for clinical requirements.
• R Package Implementation: Implements the methods as an R package integrating classifier construction and performance assessment functions.

Scientific Applications:

• Diagnostic and Prognostic Test Development: Development and validation of diagnostic, prognostic, and predictive cancer tests from high-dimensional molecular data.
• Cancer Microarray Analysis: Application to analysis of cancer microarray datasets for classifier demonstration and validation.
• Clinical Decision Support: Use of continuous scores and probability conversion to inform clinical decision-making by distinguishing borderline cases.

Methodology:

Pipeline steps explicitly include feature selection, nearest-centroid classification, computation and visualization of continuous prediction scores, conversion of scores to probabilities, and validation via Training/Test splits, Leave-One-Out Cross-Validation, and multiple random validation protocols (including an extended multiple random validation).