CancerEMC

CancerEMC implements a bagging Ensemble Meta Classifier to detect cancers early from non-invasive blood tests by integrating circulating protein biomarkers and mutations in cell-free DNA (cfDNA).

Key Features:

• Bagging Ensemble Meta Classifier: Uses a bagging Ensemble Meta Classifier approach to combine model predictions and enhance detection accuracy.
• Biomarker integration: Integrates circulating protein biomarkers and mutations in cell-free DNA (cfDNA) from blood samples for multi-analyte analysis.
• Detection modes: Performs binary cancer detection (cancer versus non-cancer) and multi-class cancer type localization from blood-derived signals.
• Class imbalance handling: Mitigates class imbalance in multi-analyte blood-test data using oversampling and adaptive synthesis techniques.
• Performance metrics: Reports binary classification accuracy of 99.17% (AUC = 0.999), multi-class localization accuracy of 74.12% (AUC = 0.938), and improved accuracy of 91.50% (AUC = 0.992) with oversampling, compared with CancerSEEK at 69.64% (AUC = 0.921).
• Validation: Evaluated on clinical blood test data and independent isolated testing datasets to assess robustness and generalizability.

Scientific Applications:

• Early non-invasive cancer detection: Detects presence of cancer from blood-derived circulating protein biomarkers and cfDNA mutations for early screening.
• Cancer versus non-cancer classification: Distinguishes cancerous from non-cancerous states in clinical blood-test samples.
• Cancer type localization: Localizes multiple cancer types (multi-class classification) from blood-derived signals.
• Addressing class imbalance in multi-analyte tests: Applies oversampling and adaptive synthesis to improve model training on imbalanced clinical datasets.
• Benchmarking and comparative evaluation: Provides comparative performance assessment against existing algorithms including CancerSEEK using accuracy and AUC metrics.

Methodology:

Computational methods explicitly include a bagging Ensemble Meta Classifier, oversampling and adaptive synthesis for class imbalance, and evaluation using accuracy and AUC on clinical and independent test datasets.