KOMPUTE

KOMPUTE imputes missing phenotype summary statistics in high-throughput model organism gene-knockout datasets to enhance functional characterization of mammalian protein-coding genes.

Key Features:

• Imputation Methodology: Employs conditional distribution properties of multivariate normal distributions to estimate association Z-scores for unmeasured phenotypes by computing their conditional expectation given observed Z-scores.
• Performance versus Matrix Completion: Demonstrates superior performance compared to singular value decomposition matrix completion in evaluations on simulated and real-world datasets.

Scientific Applications:

• Gene-Phenotype Association Studies: Enables more complete association analyses between loss-of-function genotypes from IMPC gene-knockout studies and phenotypic outcomes by imputing missing summary statistics.
• Enhancing Data Completeness: Fills gaps in datasets with widespread missingness, for example addressing ~75.6% missing association summary statistics reported in IMPC release version 16, to increase coverage for downstream analyses.

Methodology:

KOMPUTE uses the conditional expectation of multivariate normal distributions to predict missing association Z-scores based on observed phenotype Z-scores.