Discriminating Neoplastic from Nonneoplastic Tissues Using an miRNA-Based Deep Cancer Classifier

Area of Science:

• Genomics
• Bioinformatics
• Oncology

Background:

• Next-generation sequencing (NGS) generates large biological datasets, enabling advanced molecular-based disease classification.
• MicroRNAs (miRNAs) are regulatory RNA molecules quantifiable by NGS and serve as effective classification markers.

Purpose of the Study:

• To adapt and evaluate a deep cancer classifier (DCC) for differentiating neoplastic from nonneoplastic human breast and skin tissue samples.
• To compare the performance of the DCC against traditional machine learning classifiers (SVM, Random Forests) and a feature selection algorithm (cancer specificity).

Main Methods:

• Utilized comprehensive miRNA expression profiles from 1031 human breast and skin tissue samples.
• Trained and validated the DCC on 750 neoplastic and 281 nonneoplastic samples.
• Compared DCC performance (AUC, sensitivity, specificity) with support vector machine and random forests classifiers, and the cancer specificity algorithm.

Main Results:

• The DCC achieved superior performance, indicated by the highest area under the receiver operating curve (AUC), sensitivity, and specificity.
• Deep learning demonstrated advantages in handling highly heterogeneous datasets compared to other models.
• The cancer specificity algorithm identified potential biomarkers, including miR-144 and miR-375 for breast cancer, and miR-375 and miR-451 for skin cancer.

Conclusions:

• The adapted deep cancer classifier (DCC) offers a highly accurate method for distinguishing neoplastic from nonneoplastic tissues based on miRNA expression.
• Deep learning models show significant potential for improving cancer classification, especially with complex, heterogeneous data.
• Identified specific miRNAs as potential biomarkers for breast and skin cancer detection.