PreCanCell

PreCanCell predicts cell-level malignancy from single-cell transcriptomic data to classify malignant versus non-malignant cells across five cancer types including renal cell carcinoma (RCC), head and neck squamous cell carcinoma (HNSCC), melanoma, lung adenocarcinoma (LUAD), and breast cancer (BC).

Key Features:

• Differentially Expressed Genes (DEGs) Identification: Identifies DEGs between malignant and non-malignant cells in single-cell transcriptome datasets for RCC, HNSCC, melanoma, LUAD, and BC.
• k-Nearest Neighbors (k-NN) Classification: Applies k-NN classification with k = 5 using identified DEGs as features to label individual cells as malignant or non-malignant.
• Majority Voting Mechanism: Aggregates k-NN classification results via majority voting to determine final cell-level labels.
• Performance Metrics: Validated on 19 single-cell datasets with accuracy, sensitivity, specificity, balanced accuracy, and AUROC all reported above 0.8.
• Comparative Performance: Shows higher accuracy compared to CHETAH, SciBet, SCINA, scmap-cell, scmap-cluster, SingleR, and ikarus.

Scientific Applications:

• Tumor Heterogeneity Analysis: Enables identification of malignant versus non-malignant cells to study intratumoral heterogeneity at single-cell resolution.
• Precision Oncology: Supports development of personalized treatment strategies by distinguishing malignant cell populations within tumors.
• Cancer Biology Research: Facilitates basic and clinical research into cancer biology by providing cell-level malignancy annotations across multiple cancer types.

Methodology:

Extracts DEGs from training datasets for each cancer type; applies k-NN classification (k = 5) using DEGs as features; aggregates k-NN outputs by majority voting.