Related Experiment Video
Updated: Jun 16, 2026
06:05
Multiplex Immunofluorescence Combined with Spatial Image Analysis for the Clinical and Biological Assessment of the Tumor Microenvironment
Published on: June 2, 2023
Evaluating statistical models for overdispersed multiomics data: a multiplex immunofluorescence case study
Claire E Thomas1, Evertine Wesselink2, Yasutoshi Takashima3
1Public Health Sciences Division, Fred Hutchinson Cancer Center, Seattle, WA, United States.
American Journal of Epidemiology
|June 15, 2026
Summary
Ordinal logistic, linear negative binomial (NB), and quasi-Poisson models are robust for analyzing multi-omic data in cancer research. These models effectively handle overdispersed and zero-inflated data, crucial for accurate T cell subset density analysis.
Area of Science:
- Biostatistics
- Computational Biology
- Cancer Research
Background:
- Multi-omic data analysis presents significant statistical challenges, particularly in complex biological systems like colorectal cancer.
- T cell subset densities are critical biomarkers in cancer immunology, but their analysis requires robust statistical methods.
Purpose of the Study:
- To evaluate and compare the performance of seven statistical models for analyzing multiplex immunofluorescence T cell subset density data in colorectal cancer.
- To identify the most suitable models for handling overdispersed and zero-inflated multi-omic data in epidemiological studies.
Main Methods:
- A comparative analysis of seven statistical models: ordinal logistic regression, Poisson, quasi-Poisson, quadratic negative binomial (NB), linear NB, zero-inflated NB (ZINB), and hurdle NB.
- Assessment of model performance using a large dataset (1235 cases) of colorectal cancer multiplex immunofluorescence data.
- Simulation studies to evaluate type I error rates and statistical power across various data distributions and zero proportions.
Main Results:
- Poisson and NB-based models (quadratic, ZINB, hurdle) exhibited inflated false-positive rates.
- Ordinal logistic regression and linear NB models showed no significant inflation of false positives.
- Ordinal logistic, linear NB, and quasi-Poisson models demonstrated superior or near-optimal statistical power across different data scenarios, including varying zero proportions and dispersion levels.
Conclusions:
- Ordinal logistic, linear NB, and quasi-Poisson models are recommended as robust and effective statistical tools for analyzing overdispersed, right-skewed multi-omic data with a substantial proportion of zero counts.
- These models provide reliable effect estimates, particularly for weaker exposures, and maintain appropriate statistical properties for epidemiological research in cancer.
- The findings offer valuable guidance for researchers in bioinformatics and cancer epidemiology dealing with complex high-dimensional biological data.
