Precision cancer classification using liquid biopsy and advanced machine learning techniques.

Area of Science:

• Oncology
• Biotechnology
• Bioinformatics

Background:

• Cancer poses a significant global health challenge, underscoring the need for early detection methods.
• Liquid biopsy, analyzing circulating cell-free DNA (cfDNA/ctDNA) and biomarkers, offers a promising avenue for non-invasive cancer diagnosis.
• Timely cancer detection is crucial for improving patient survival rates and enabling effective treatment.

Purpose of the Study:

• To develop and validate a machine learning system for early cancer detection and classification using plasma cfDNA/ctDNA mutations and protein biomarkers.
• To enhance the efficiency and accuracy of cancer detection by employing advanced feature selection and classification techniques.

Main Methods:

• Utilized correlation coefficient and mutual information for feature selection, reducing data dimensionality by 60% using XGBoost feature importance.
• Employed Light Gradient Boosting Machine (LGBM) for classification, optimizing hyperparameters via random search.
• Ensembled tenfold cross-validated LGBM models, weighted by balanced accuracy, for final predictions.

Main Results:

• Achieved 99.45% accuracy and 99.95% AUC for cancer presence detection.
• Attained 93.94% accuracy and 97.81% AUC for cancer-type classification.
• Demonstrated a significant reduction in dataset dimensionality while maintaining high predictive performance.

Conclusions:

• The proposed liquid biopsy system demonstrates high efficacy in early cancer detection and classification.
• This methodology holds potential for improving patient outcomes through timely and accurate cancer diagnosis.
• The integration of cfDNA/ctDNA analysis and protein biomarkers offers a powerful tool in the fight against cancer.