g3mclass

g3mclass applies Gaussian mixture modeling to classify molecular assay data for biomarker qualification and diagnostic assessment.

Key Features:

• Gaussian Mixture Modeling (GMM): Implements Gaussian mixture modeling for distribution-based classification of molecular assay measurements.
• Semi-constrained EM algorithm: Uses a semi-constrained expectation-maximization (EM) algorithm for parameter estimation and automated classification.
• Multiclass probabilistic classifiers: Automates application of validated multiclass classifiers and yields probabilistic class assignments.
• Assay support: Processes both single analyte tests and multiplexing assays.
• Demonstrated datasets and biomarkers: Validated on clinical and gene expression datasets including ERBB2, ESR1, and PGR and applied to breast cancer diagnostic challenges.
• Accuracy, robustness, and scalability: Produces outputs intended to be accurate and robust across diverse datasets and scalable to larger data volumes.
• Interpretability: Produces interpretable probabilistic classification results to support diagnostic decision-making.
• Integration with ML/AI: Uses probabilistic modeling that is adaptable for integration with machine learning and artificial intelligence frameworks to support personalized medicine and therapeutic interventions.

Scientific Applications:

• Biomarker qualification: Classifies biomarkers to support qualification and validation workflows.
• Diagnostic assessment: Supports diagnostic classification to reduce overdiagnosis, underdiagnosis, and equivocal results, exemplified in breast cancer diagnostics.
• Companion diagnostics and regulatory evaluation: Provides classification outputs useful for companion diagnostic development and evaluation by healthcare regulators.
• Early disease assessment and personalized medicine: Enables early assessment of human diseases and contributes to personalized medicine strategies.
• Therapy monitoring and intervention design: Supports monitoring of therapy responses and the design of targeted therapeutic interventions through biomarker classification.

Methodology:

Applies Gaussian mixture modeling (GMM) with a semi-constrained expectation-maximization (EM) algorithm to produce validated multiclass probabilistic classifiers for single-analyte and multiplexing assay data.