countfitteR: efficient selection of count distributions to assess DNA damage

Area of Science:

• Biomedical Imaging
• Computational Biology
• Statistical Genetics

Background:

• Counting DNA double-strand breaks (DSBs) as foci is crucial for personalized medicine, cancer research, and drug efficacy evaluation.
• The standard Poisson distribution assumption for count data can be inaccurate, leading to biased results due to zero-inflation or overdispersion.
• Selecting appropriate statistical models (e.g., Negative Binomial, Zero-Inflated models) is critical but challenging for non-Poisson distributed data.

Purpose of the Study:

• To develop an automated and objective software tool for selecting appropriate statistical distribution models for count data.
• To simplify and enhance the analysis of DNA double-strand break foci count data.
• To provide statistically verifiable parameter estimations and confidence intervals.

Main Methods:

• Developed countfitteR, an R package utilizing a Bayesian approach for distribution model selection.
• Integrated countfitteR with the shiny web application framework for interactive data analysis.
• Applied a Bayesian framework for robust model comparison and selection.

Main Results:

• countfitteR demonstrated superior or equal statistical performance compared to traditional two-step methods, achieving up to 98% power.
• The software provided valuable insights even when traditional methods yielded no results.
• Analysis of DNA damage data revealed the Negative Binomial distribution as most frequent, followed by the Poisson distribution.

Conclusions:

• countfitteR automates distribution model selection, enhancing objectivity and statistical rigor in count data analysis.
• The software is valuable for analyzing foci in biomedical imaging and other fields with non-Poisson count data.
• Facilitates reliable parameter estimation and confidence intervals for complex biological datasets.