CancerCSP

CancerCSP predicts the stage of clear cell renal cell carcinoma (ccRCC) from RNA-Seq gene expression data (RSEM) to identify stage-associated biomarkers.

Key Features:

• Input data: Uses RNA-Seq by Expectation Maximization (RSEM) values as the gene expression input.
• Cohort analyzed: Analyzed gene expression profiles from 523 ccRCC samples.
• Single-gene threshold classifier (NR3C2): A threshold-based method using NR3C2 achieved 71.12% accuracy with an ROC of 0.67.
• Eight-gene multi-gene panel: A combinatorial panel of eight genes (underexpressed: NR3C2, ENAM, DNASE1L3, FRMPD2; overexpressed: PLEKHA9, MAP6D1, SMPD4, C11orf73) achieved 70.19% accuracy with an ROC of 0.74 on a validation dataset.
• State-of-the-art models (64 genes): Models developed with state-of-the-art techniques on a 64-gene set reached a maximum accuracy of 72.64% with an ROC of 0.81 on a validation dataset.
• Hallmark gene subset (38 genes): A 38-gene subset involved in cancer hallmark biological processes produced similar accuracy to the 64-gene models.
• Validation metrics: Performance reported using accuracy and ROC values on validation datasets.
• Gender-specific modeling consideration: The study indicated a potential necessity for gender-specific models to improve stage classification accuracy.

Scientific Applications:

• Stage prediction: Predicts early versus late stage classification of ccRCC from gene expression data.
• Biomarker identification: Identifies genes associated with ccRCC stage, including single-gene and multi-gene signatures.
• Panel evaluation: Assesses performance of single-gene, eight-gene, 38-gene, and 64-gene panels for stage classification.
• Biological process analysis: Evaluates genes involved in cancer hallmark biological processes for their role in staging.
• Model stratification research: Supports investigation into gender-specific models to refine stage classification.

Methodology:

Analysis of RSEM gene expression from 523 ccRCC samples using a threshold-based method and multi-gene combinations, development of models using state-of-the-art techniques on 64-gene and 38-gene subsets, and evaluation on validation datasets using accuracy and ROC metrics.