GRRANN

GRRANN predicts phenotypes from transcriptomic (gene expression) data by integrating a parsed version of the STRING DB biological network into a group-wise regularized artificial neural network for clinical trial applications.

Key Features:

• Biological Network Integration: Incorporates regulatory interactions from a parsed version of the STRING DB to encode gene-regulatory mechanisms.
• Regularization for Sparsity and Reproducibility: Applies group-wise regularization to shrink gene sets based on active upstream regulatory mechanisms, promoting sparsity and reproducibility.
• Robustness and Generalizability: Integration of biological networks improves robustness and generalizability across independent test sets in clinical datasets.
• Interpretability of Biological Signatures: Regularization increases interpretability of derived biological signatures used for phenotype prediction.

Scientific Applications:

• Patient stratification: Stratifying patient subpopulations by treatment response or likelihood of adverse reactions using genome-wide expression data.
• Clinical outcome prediction: Predicting acute rejection in kidney transplantation and response to Infliximab in ulcerative colitis using clinical trial transcriptomic data.
• Benchmarking: Comparative evaluation against regression models, support vector machines, and artificial neural networks.

Methodology:

Uses transcriptomic measurements from clinical trials; parses STRING DB to obtain gene-regulatory interactions; constructs a group-wise regularized artificial neural network that integrates the biological network; applies group-wise regularization to shrink gene sets; benchmarks the model against regression models, support vector machines, and artificial neural networks using clinical trial outcomes such as acute rejection in kidney transplantation and response to Infliximab in ulcerative colitis.