PrismEXP

PrismEXP stratifies RNA-seq co-expression data by tissue and cell type to improve prediction of gene annotations such as pathway membership, Gene Ontology terms, and phenotypic associations in human and mouse.

Key Features:

• Stratification for Specificity: PrismEXP stratifies uniformly aligned RNA-seq data by tissue and cell type to capture context-specific gene co-expression.
• Improved Predictive Performance: By partitioning global gene-gene co-expression matrices into tissue- and cell type-specific matrices, PrismEXP enhances accuracy for predicting pathway membership, Gene Ontology terms, and phenotypic associations in human and mouse.
• Cross-Domain Transfer Learning: Training on one type of gene annotation can facilitate prediction of annotations in other domains through transfer of learned patterns.
• Integration with Machine Learning: PrismEXP enhances unsupervised machine learning analyses to provide deeper insights into the roles of understudied genes and proteins.

Scientific Applications:

• Context-specific gene function prediction: Inferring gene functions and interactions within specific tissues and cell types.
• Annotation prediction across domains: Predicting pathway membership, Gene Ontology terms, and phenotypic associations from co-expression data.
• Disease mechanism and target discovery: Supporting exploration of genetic networks, disease mechanisms, and potential therapeutic targets.
• Discovery of novel gene functions: Identifying gene functions and interactions that may be overlooked by global, cross-tissue correlation approaches.

Methodology:

PrismEXP uses uniformly aligned RNA-seq data from ARCHS4, partitions global gene-gene co-expression matrices into tissue- and cell type-specific matrices, applies a stratified approach to generate refined predictions across annotation domains, enables transfer learning between annotation domains, and integrates with unsupervised machine learning methods.