Machine-learning diagnostics of breast cancer using piRNA biomarkers

Area of Science:

• Biomolecular Sciences
• Computational Biology
• Oncology

Background:

• Small non-coding RNAs (sncRNAs) are implicated in cancer development.
• PIWI-interacting RNAs (piRNAs), a class of small ncRNAs, play a role in cancer initiation.
• piRNAs are being explored as novel, noninvasive diagnostic biomarkers for breast cancer.

Purpose of the Study:

• To develop computational methods for breast cancer prediction using piRNA attributes.
• To assess the diagnostic potential of piRNAs for breast cancer.

Main Methods:

• Generated piRNA sequence descriptors.
• Applied machine learning classifiers (Logistic Regression, SMO, Random Forest, LMT) in WEKA.
• Utilized Shapley additive explanations (SHAP) for descriptor relevance analysis.
• Validated models on an independent dataset.

Main Results:

• Logistic Regression achieved 90.7% accuracy in breast cancer prediction.
• Sequential Minimal Optimization (SMO) and Logistic Model Tree (LMT) showed high accuracy (89.7% and 85.65%, respectively).
• Identified key piRNA descriptors contributing to prediction accuracy.

Conclusions:

• Machine learning-based piRNA classifiers are effective for breast cancer diagnosis.
• This study supports piRNAs as valuable biomarkers in cancer diagnostics.
• Further research is needed to validate clinical applicability.