HNSCPred

HNSCPred predicts Head and Neck Squamous Cell Carcinoma presence and HPV status from single-cell transcriptomics using machine learning and deep learning for diagnostic classification.

Key Features:

• Machine learning and deep learning models: Employs classification models including an Artificial Neural Network (ANN) to distinguish HNSCC from normal tissue and to classify HPV-positive (HPV+) versus HPV-negative (HPV-) samples.
• Dataset utilization: Uses the GSE181919 single-cell transcriptomics dataset comprising 20 primary HNSCC samples and 9 normal tissue samples with representation of both HPV+ and HPV- cases.
• Feature selection: Applies minimum Redundancy Maximum Relevance (mRMR) to reduce features to 100 informative genes.
• Gene Ontology analysis: Performs GO enrichment analysis on selected genes, identifying predominant involvement in binding and catalytic activities.
• Performance metrics: The ANN model achieved an AUROC of 0.91 for HNSCC versus normal classification and an AUROC of 0.83 for HPV status classification on the validation set.

Scientific Applications:

• Diagnostic classification: Classifies single-cell transcriptomic profiles to detect HNSCC and determine HPV status.
• Biomarker prioritization: Identifies a 100-gene subset for downstream biomarker validation and functional studies.
• Molecular characterization: Enables single-cell–level analysis of gene expression patterns associated with HNSCC and HPV-related differences.

Methodology:

Models were trained on 80% of the GSE181919 single-cell transcriptomics data and validated on the remaining 20%; feature selection used mRMR to select 100 genes, GO enrichment was performed on selected genes, and classification models including an ANN were evaluated using AUROC for HNSCC versus normal and for HPV status.