Bayesian functional data analysis over dependent regions and its application for identification of differentially...

Area of Science:

• Statistics
• Bioinformatics
• Genomics

Background:

• Analyzing extremely long sequences in biological data presents challenges due to potential dependencies between adjacent segments.
• Existing methods may struggle to capture the nuances of functional patterns within and across these segments.

Purpose of the Study:

• To develop a Bayesian functional data analysis framework for long sequential observations.
• To address the dependence structure between non-independent windows within a sequence.
• To identify functional differences between groups of individuals at specific windows.

Main Methods:

• Utilized Bayesian smoothing splines for estimating individual functional patterns within each window.
• Developed transition models for parameters to capture the dependence structure between neighboring windows.
• Employed Bayes factors derived from Markov Chain Monte Carlo (MCMC) samples for group comparisons.

Main Results:

• The proposed method demonstrated effectiveness in simulation studies.
• Successfully applied to identify differentially methylated genetic regions in TCGA lung adenocarcinoma data.
• Quantified functional differences between groups at each window using statistical evidence.

Conclusions:

• The proposed Bayesian functional data analysis method is suitable for long sequential data with inter-window dependencies.
• The approach provides a robust framework for identifying group-specific functional differences, particularly in genomic applications.
• This method enhances the analysis of complex biological sequences, such as those found in cancer genomics.