A probabilistic model to predict clinical phenotypic traits from genome sequencing.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Large-scale genetic screening is advancing, but interpreting variant genotypes remains challenging.
• Individuals often seek to interpret their genetic information despite interpretation difficulties.
• Projects like the Personal Genome Project (PGP) generate vast amounts of genomic and phenotypic data.

Purpose of the Study:

• To develop a computational model for predicting clinical phenotypes from genetic data.
• To integrate diverse data sources for probabilistic phenotype prediction.
• To assess the accuracy of phenotype prediction models.

Main Methods:

• Designed a Bayesian probabilistic model to predict dichotomous phenotypes.
• Applied the model to a cohort from the Personal Genome Project (PGP).
• Evaluated model performance using area-under-the-ROC curve (AUC) and statistical significance tests.
• Participated in a Critical Assessment of Genome Interpretation (CAGI) blinded prediction experiment.

Main Results:

• The Bayesian model accurately predicted phenotypes such as Gilbert syndrome, Graves' disease, and blood groups in the PGP cohort.
• 26% of PGP phenotypes were predicted with AUC > 0.7, with 15.8% being statistically significant.
• The model achieved the most accurate prediction in a CAGI blinded experiment matching genomes to phenotypes.

Conclusions:

• The developed Bayesian model demonstrates accurate prediction of clinical phenotypes from genetic information.
• While not yet diagnostically sufficient, the model's performance is promising and expected to improve with more data.
• This approach represents a significant step towards integrating genetic data for personalized health insights.