surviveAI

surviveAI predicts long-term survival outcomes of cancer patients using somatic RNA-Seq expression data.

Key Features:

• Input Data: Uses somatic RNA-Seq expression data from The Cancer Genome Atlas (TCGA) spanning 16 cancer-type cohorts.
• Model Development: Builds Random Forest models optimized via parameter grid search and backward feature elimination for dimensionality reduction.
• Validation: Performs external validation using Clinical Proteomic Tumor Analysis Consortium v3 (CPTAC3) data.
• Performance Metrics: Reports that five models achieved area under the receiver operating characteristic curve (AUC-ROC) greater than 80%.
• Pathway and Network Analysis: Identifies pathways and networks involved in tumorigenesis and organ development and highlights genes DMBT1, IL11, HOXB6, TRIB3, and PIM1, notably in the TCGA-KIRP (renal cancer) cohort.
• Implementation: Implements machine learning using the Python scikit-learn package.

Scientific Applications:

• Prognosis Prediction: Stratifies patient survival risk and supports prediction of long-term outcomes from RNA-Seq expression signatures.
• Therapeutic Target Identification: Identifies candidate prognostic genes and pathways (e.g., DMBT1, IL11, HOXB6, TRIB3, PIM1) for potential therapeutic targeting.

Methodology:

Modeling used Random Forest implemented with Python scikit-learn, optimized by parameter grid search and backward feature elimination on somatic RNA-Seq data from TCGA (16 cohorts), with external validation on CPTAC3 and downstream pathway and network analyses to identify implicated genes and processes.