CancerLSP

CancerLSP predicts clinical stage and discriminates tumor versus normal samples in liver hepatocellular carcinoma (LIHC) using genomic and epigenomic profiles from The Cancer Genome Atlas (TCGA).

Key Features:

• Data Utilization: Uses TCGA LIHC data comprising 173 early-stage, 177 late-stage, and 50 adjacent normal tissue samples with measurements for 60,483 RNA transcripts and 485,577 methylation CpG sites.
• Feature Selection: Employs advanced feature selection techniques to identify transcriptomic expressions and methylation-based features crucial for classification.
• Machine Learning Models: Implements various machine learning algorithms to classify early versus late stages and cancer versus normal, including a Naive Bayes model using 51 combined features (21 CpG methylation sites and 30 RNA transcripts) that achieved MCC 0.58 and 78.87% accuracy for early versus late classification.
• High Precision Classification: Models based on 5 RNA transcripts and 5 CpG sites classify LIHC versus normal samples with 96-98% accuracy and an AUC of 0.99.
• Multiclass Classification: Supports multiclass classification into normal, early-stage, and late-stage categories with 76.54% accuracy and an AUC of 0.86.

Scientific Applications:

• Clinical staging and prognosis: Classifies LIHC samples into early or late stages and discriminates tumor from normal to inform therapeutic decisions and prognostic assessment.
• Molecular marker identification: Identifies transcriptomic and methylation features associated with LIHC progression to support research into disease mechanisms and stratified treatment approaches.

Methodology:

Performs analysis of differentially expressed RNA transcripts and methylated CpG sites from TCGA, applies feature selection techniques, and integrates genomic and epigenomic data into machine learning models, including Naive Bayes.