PhenoExam: gene set analyses through integration of different phenotype databases

Alejandro Cisterna, Aurora González-Vidal, Daniel Ruiz, Jordi Ortiz, Alicia Gómez-Pascual, Zhongbo Chen, Mike A. Nalls, Faraz Faghri, John Hardy, Irene Díez, Paolo Maietta, Sara Álvarez, Mina Ryten, Juan A. BotíaBMC Bioinformatics 2022 · DOI

TL;DR

PhenoExam is a new R package and web tool that integrates multiple phenotype and disease databases to perform comprehensive gene set enrichment analysis. It helps researchers identify gene-phenotype associations, compare similarities between gene sets, and discover differential phenotypes, offering a more nuanced understanding of genetic diseases.

What Problem Does This Paper Address?

Identifying the genetic basis of diseases is a cornerstone of clinical genetics. Diseases are often characterized by a set of clinical symptoms and signs that form a recognizable phenotype. While environmental factors play a role, genetic variations are frequently implicated in inherited disorders. However, linking specific genes to complex phenotypes and diseases remains a significant challenge. Existing tools often rely on a single phenotype database, limiting their scope. Furthermore, comparing gene sets to understand subtle phenotypic differences between similar diseases or to validate predicted disease genes has been cumbersome. This paper addresses the need for a more integrated and flexible approach to phenotype-based gene set analysis.

Methods

PhenoExam offers a multi-pronged approach to phenotype analysis:

  • Phenotype and Disease Enrichment Analysis: This core function identifies phenotypic terms that are significantly enriched within a given set of genes. It leverages a hypergeometric distribution to model enrichment probabilities and offers adjustments for multiple testing (Bonferroni, FDR).
  • Phenotype Similarity Measurement: PhenoExam quantifies the similarity between two gene sets (G and G') using the Phenotypic Overlap Ratio (POR). Users can choose between the Jaccard index or the Forbes similarity coefficient. The significance of the POR is assessed through randomization.
  • Differential Phenotype Analysis: This feature identifies phenotypes that are significantly enriched in one gene set but not the other, allowing for the detection of exclusive or distinguishing phenotypic terms.
  • Database Integration: A key innovation is PhenoExam's ability to integrate data from numerous curated databases, including HPO, MGI, OMIM, UniProt, CTD, Orphanet, ClinGen, Genomics England PanelApp, CGI, PsyGeNET, and CRISPRBrain. This allows for a more comprehensive analysis across different organisms (human and mouse) and data types.

Datasets

PhenoExam integrates data from the following databases:

  • Human Phenotype Ontology (HPO): Standardized human phenotypic terms.
  • Mouse Genome Database (MGD): Comprehensive data on mouse genes, phenotypes, and models of human disease.
  • Online Mendelian Inheritance in Man (OMIM): Catalog of human genes, genetic diseases, and traits.
  • UniProt: Protein knowledgebase.
  • Comparative Toxicogenomics Database (CTD): Gene-disease associations.
  • Orphanet: Database of rare diseases and orphan drugs.
  • Clinical Genome Resource (ClinGen): Gene-disease associations.
  • Genomics England PanelApp: Diagnostic gene panels.
  • Cancer Genome Interpreter (CGI): Gene-disease (cancer) associations.
  • PsyGeNET: Gene-disease associations for psychiatric disorders.
  • CRISPRBrain: Genome-wide CRISPR interference and activation screens in human neurons.

The paper does not introduce new benchmark datasets but rather utilizes and integrates existing ones.

Key Insights and Ideas

  • The Power of Integration: The most significant insight is the value of integrating multiple phenotype and disease databases. This approach provides a richer, more context-aware analysis than relying on a single source, leading to more robust gene-phenotype associations.
  • Distinguishing Subtle Differences: PhenoExam's ability to detect differential phenotypes is crucial for differentiating between gene sets or diseases that appear phenotypically similar at a high level but have underlying distinctions. This is vital for precise diagnostics and understanding disease mechanisms.
  • Validation of Computational Predictions: The tool's utility in validating computationally predicted disease genes is a major advantage. By comparing predicted genes to known gene sets, researchers can assess the biological plausibility of their predictions.
  • Flexibility for Different Users: PhenoExam is designed to be accessible to clinicians, computational biologists, and geneticists, offering both a user-friendly web interface and a more flexible R package for advanced users.

Key Results

  • Accurate Phenotype Enrichment: PhenoExam demonstrates sensitive and accurate phenotype enrichment analyses.
  • Type I Error Control: Simulations show that PhenoExam generally maintains type I error under control, though this can be influenced by the number of average disease terms per gene and the gene set size. Recommendations are provided for optimal gene set sizes for different databases.
  • Distinguishing Similar Diseases: The tool successfully distinguished between Parkinson's disease (PD) and early-onset dystonia (EOD), highlighting both shared and differential phenotypic terms.
  • Validating Epilepsy Genes: PhenoExam was used to validate computationally predicted epilepsy genes, showing they recapitulate phenotype terms of known epilepsy genes, thus supporting the prediction methodology.
  • POR Sensitivity: The Phenotypic Overlap Ratio (POR) was shown to be sensitive to small changes (as low as 1%) in gene composition, demonstrating its ability to detect subtle differences between gene sets.

Limitations and Future Work

  • Database Dependency: The accuracy and comprehensiveness of PhenoExam's analyses are inherently dependent on the quality and completeness of the integrated databases. Users need some basic knowledge of the diseases and databases to interpret results appropriately.
  • Type I Error in Specific Databases: While generally controlled, type I error can be more challenging to manage with certain databases (e.g., Genomics England Panel App, Orphanet) for smaller gene sets. The authors suggest using larger gene sets for these databases.
  • Future Work: The authors suggest that PhenoExam can be extended to incorporate more databases and potentially new analysis types. The interactive plots and tutorials aim to foster visualization and interpretation, suggesting a continued focus on user-friendliness and interpretability. The tool's ability to switch between gene and phenotype spaces also hints at potential for further exploration of these relationships.

Summary updated

BibTeX
@article{cisterna2022phenoexam,
  author = {Alejandro Cisterna and Aurora González-Vidal and Daniel Ruiz and Jordi Ortiz and Alicia Gómez-Pascual and Zhongbo Chen and Mike A. Nalls and Faraz Faghri and John Hardy and Irene Díez and Paolo Maietta and Sara Álvarez and Mina Ryten and Juan A. Botía},
  title = {{PhenoExam: gene set analyses through integration of different phenotype databases}},
  journal = {BMC Bioinformatics},
  year = {2022},
  doi = {10.1186/s12859-022-05122-x},
}
Figure