Pheno-Ranker

Pheno-Ranker compares phenotypic data stored in GA4GH standards, enabling individual-level comparisons and statistical significance analysis for disease association studies, patient stratification, and genotype-phenotype correlation analysis.

Key Features:

• Multi-format input: Accepts JSON/YAML data exchange formats from Beacon v2 and Phenopackets v2 data models, as well as any data structure encoded in JSON, YAML, or CSV formats.
• Hierarchical data transformation: Flattens hierarchical data structures to one dimension and transforms them through one-hot encoding for efficient comparison.
• Pairwise comparison: Performs efficient pairwise (all-to-all) comparisons within cohorts or matches a patient's profile against cohorts.
• Customizable comparison parameters: Allows users to refine comparisons by including or excluding terms and applying weights to variables.
• Statistical significance testing: Provides Z-scores and p-values for statistical significance of comparisons.
• Output for downstream analysis: Generates text files that can be analyzed using unsupervised learning techniques such as clustering or multidimensional scaling (MDS), and with graph analytics.
• Additional utilities: Includes utilities for CSV import, data simulation, summary statistics plotting, and QR code generation.

Scientific Applications:

• Disease association studies: Enables comparison of phenotypic data to identify associations between phenotypes and diseases.
• Patient stratification: Supports grouping of patients based on phenotypic similarities for clinical research.
• Genotype-phenotype correlation analysis: Facilitates analysis of relationships between genetic variants and phenotypic manifestations.
• Cohort comparison: Allows matching of individual patient profiles against existing cohorts for clinical research applications.
• Semantic similarity analysis: Performs analysis of phenotypic data in Beacon v2 and Phenopackets v2 formats, extendable to other data types and a wide range of variables beyond HPO or OMIM terms while preserving full context.

Methodology:

Hierarchical data structures are flattened to one dimension and transformed through one-hot encoding. Pairwise comparisons are performed within cohorts or for matching a patient's profile in cohorts, with statistical significance obtained through Z-scores and p-values. Performance was validated with simulated and synthetic data, and a real data use case from the PRECISESADS study demonstrates practical utility in clinical research.