NCIS

NCIS integrates molecular interaction networks with weighted co-clustering using semi-nonnegative matrix tri-factorization to simultaneously group genes and samples from high-throughput gene expression data for identification of clinically relevant cancer subtypes.

Key Features:

• Weighted co-clustering algorithm: Employs semi-nonnegative matrix tri-factorization to co-cluster genes and samples.
• Gene weighting by network impact: Assigns weights to genes based on their impact within molecular interaction networks so influential genes have greater effect on clustering.
• Network integration: Incorporates molecular interaction network information into the clustering process.
• High-dimensional data support: Operates on high-throughput, high-dimensional gene expression datasets.
• Noise robustness: Demonstrates higher accuracy in handling noise within the data.
• Comparative performance: Separates patient samples into clinically relevant subtypes more effectively than consensus hierarchical clustering.
• Validation datasets: Evaluated on simulated datasets and patient samples from The Cancer Genome Atlas (TCGA).

Scientific Applications:

• Cancer subtype identification: Identification of clinically distinct cancer subtypes from gene expression data using network-aware co-clustering.
• Tumor heterogeneity analysis: Disentangling tumor heterogeneity by simultaneous clustering of genes and samples.
• Specific cancer studies: Applied to studies of breast cancer and glioblastoma multiforme.
• Clinical stratification: Informing stratification of patients for more precise or personalized therapeutic strategies based on molecular subtypes.

Methodology:

Weighted co-clustering using semi-nonnegative matrix tri-factorization with gene weights derived from molecular interaction network impact, applied to high-throughput gene expression data and evaluated on simulated datasets and TCGA samples with comparisons to consensus hierarchical clustering.