Cox-nnet

Cox-nnet applies artificial neural networks to predict patient prognosis from high-throughput omics and genomics data for survival analysis.

Key Features:

• Neural network survival modeling: Uses artificial neural network methodologies to model time-to-event outcomes and predict prognosis.
• High-throughput omics support: Supports analysis of high-throughput omics and genomics data for survival prediction.
• Version 2.0 efficiency: Cox-nnet v2.0 reports up to a 32-fold reduction in training time on a kidney transplantation dataset of 10,000 samples.
• Large-scale dataset suitability: Applicable to large-scale population datasets, including electronic medical records (EMR).
• Interpretability: Implements permutation-based feature importance scores and provides directionality information for feature coefficients.
• Improved predictive performance: Demonstrates statistically significant accuracy improvements (P<0.05) versus Cox proportional hazards (Cox-PH) regression on a kidney transplantation dataset and outperforms on the SUPPORT dataset (8,000 samples).

Scientific Applications:

• Prognosis prediction from omics: Predicts patient outcomes using high-throughput omics and genomics data.
• Survival analysis in transplantation: Applied to kidney transplantation datasets for time-to-event outcome modeling and performance benchmarking.
• EMR-based outcome modeling: Applied to large-scale electronic medical record datasets for population-level survival prediction.

Methodology:

Uses artificial neural network-based survival modeling, permutation-based feature importance scoring, extraction of feature coefficient directionality, and comparative evaluation against Cox-PH regression with reported training-time measurements (32-fold reduction) on a 10,000-sample kidney transplantation dataset and performance evaluation on the 8,000-sample SUPPORT dataset.