An information theoretic approach to pedigree reconstruction

Area of Science:

• Computational Biology
• Bioinformatics
• Statistical Genetics

Background:

• Biological systems exhibit complex network structures at cellular and population levels.
• Advances in genetic screening enable network reconstruction, but statistical inference remains challenging.
• Traditional statistical methods are often unsuitable for nonstandard network data structures.

Purpose of the Study:

• To develop a robust statistical inference method for biological network reconstruction.
• To apply an information-theoretic approach, specifically the Minimum Description Length (MDL) principle, within a Bayesian framework.
• To evaluate the MDL approach's effectiveness in pedigree reconstruction, particularly concerning genotyping errors.

Main Methods:

• A Bayesian inference model was constructed on the space of network structures.
• The Minimum Description Length (MDL) principle was employed, focusing on information content rather than probabilistic models.
• The MDL approach was applied to pedigree reconstruction using genotypic data and compared against COLONY-2 via a simulation study.

Main Results:

• The MDL approach provides objective control for model complexity in network inference.
• Unlike COLONY-2, the MDL method does not require prior genotyping error rate estimates.
• The MDL approach accurately adjusts for genotyping errors, demonstrating superior performance in pedigree reconstruction simulations.

Conclusions:

• The MDL-based Bayesian inference model offers a powerful and objective tool for analyzing biological network structures.
• This method overcomes limitations of traditional statistical approaches and improves accuracy in tasks like pedigree reconstruction.
• The MDL approach's ability to handle genotyping errors without prior estimation represents a significant advancement in genetic data analysis.