Prioritizing and characterizing functionally relevant genes across human tissues.

Area of Science:

• Genomics
• Computational Biology
• Systems Biology

Background:

• Identifying genes critical for specific tissue functions is challenging, hindering mechanistic understanding of genotype-phenotype links.
• Current methods often rely solely on tissue-specific expression, which may not fully capture a gene's functional relevance.

Purpose of the Study:

• To develop and validate FUGUE, a novel machine learning model for predicting tissue-relevant genes across 30 human tissues.
• To integrate transcriptional and network features for improved gene prioritization.

Main Methods:

• Developed FUGUE, a machine learning model combining transcriptional and network data.
• Trained and validated the model across 30 human tissues using cross-validation.
• Assessed performance on independent datasets and compared it with conventional metrics.

Main Results:

• FUGUE achieved an average cross-validation auROC of 0.86 and auPRC of 0.50.
• The model accurately distinguished tissue- or cell type-specific genes, outperforming expression-based metrics.
• Tissue-relevant genes showed genomic clustering within topologically associated domains and enrichment in cancer-related differentially expressed genes.

Conclusions:

• FUGUE effectively predicts tissue-relevant genes, advancing the understanding of genotype-phenotype relationships.
• The identified genes have implications for understanding tissue development, function, and disease, particularly cancer.
• Prioritized gene lists and open-source software are provided for broader research applications.