CancerTSP

CancerTSP predicts and classifies Papillary Thyroid Carcinoma (PTC) samples using RNA-Seq FPKM expression data to distinguish early versus late stages and cancerous versus normal tissues.

Key Features:

• Stage Prediction Module: Predicts early versus late-stage PTC from RNA-Seq FPKM expression profiles.
• Cancer vs. Normal Prediction Module: Classifies samples as cancerous or normal using selected RNA transcripts.
• Data Analysis Module: Analyzes expression of specific RNA transcripts across early and late PTC stages.
• Dataset: Uses RNA-Seq FPKM profiles from The Cancer Genome Atlas (TCGA) THCA dataset comprising 500 cancer and 58 normal samples.
• Feature Selection: Applies various feature selection techniques to identify discriminative RNA transcripts.
• Machine Learning Classification: Employs machine learning algorithms to classify samples based on selected features.
• Single-Gene Marker: Identified DCN as a single-gene marker with AUROC of 0.66 for distinguishing early from late-stage samples.
• 36-Transcript Panel: A panel of 36 RNA transcripts achieved an F1 score of 0.75 and an AUROC of 0.73 (95% CI: 0.62–0.84) on the validation dataset and correctly predicted 75% of samples in an external validation set.
• Multiclass Model: Multiclass model classifies normal, early, and late-stage samples with AUROCs of 0.95, 0.76, and 0.72 respectively on the validation dataset.
• Five-Transcript Cancer vs Normal Panel: A panel of five protein-coding transcripts segregated cancer from normal samples with an F1 score of 0.97 and an AUROC of 0.99 (95% CI: 0.91–1).

Scientific Applications:

• Thyroid Cancer Diagnosis: Supports diagnostic discrimination of PTC stages and cancer versus normal using RNA-Seq expression markers such as DCN and transcript panels.
• Research on Thyroid Carcinoma Progression: Enables investigation of gene expression changes associated with progression from early to late PTC stages.
• Biomarker Discovery and Validation: Facilitates discovery and external validation of single-gene and multi-transcript biomarkers with reported AUROC and F1 performance metrics.

Methodology:

Analyzed RNA-Seq FPKM expression profiles from TCGA THCA (500 cancer, 58 normal), applied various feature selection techniques to identify discriminative RNA transcripts, and used machine learning algorithms to classify samples.