Haplotype Frequency Inference From Pooled Genetic Data With a Latent Multinomial Model.

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Haplotype data offer refined genetic information compared to individual markers.
• Large-scale genetic studies often yield pooled data, necessitating robust inference methods.
• Existing normal approximation methods for pooled data inference can be unreliable, especially with near-singular covariance matrices.

Purpose of the Study:

• To develop and present exact computational methods for inferring haplotype frequencies from pooled genetic data.
• To address the limitations of current approximate methods that fail on real-world datasets.
• To provide a scalable and accurate alternative for analyzing large-scale genetic association studies.

Main Methods:

• Proposed two exact inference methods based on a latent multinomial model.
• Utilized integer combinations of latent, unobserved haplotype counts to model pooled results.
• Developed a latent count sampling method via Markov bases with approximately linear runtime relative to pool size.

Main Results:

• The proposed exact methods demonstrate superior accuracy compared to existing approximate methods.
• Validation performed on both synthetic datasets and real haplotype data from the 1000 Genomes Project.
• The methods successfully applied to time-series pooled genetic data, showing potential for complex hierarchical models.

Conclusions:

• Exact methods provide more reliable haplotype frequency inference from pooled genetic data than traditional approximations.
• The developed methods are computationally efficient and accurate, suitable for large-scale genetic studies.
• The approach is adaptable to more complex data structures, including spatiotemporal genetic analyses.