Estimating pseudocounts and fold changes for digital expression measurements.

Area of Science:

• Bioinformatics
• Computational Biology
• Statistical Genetics

Background:

• Count-based high-throughput experiments, such as RNA-sequencing, often encounter a zero-frequency problem when calculating fold changes.
• Current methods to address this issue involve adding pseudocounts, but their magnitude and introduction stage are arbitrary and can lead to biased results.
• Fold changes, in a strict sense, are not directly computable quantities but require statistical inference due to experimental stochasticity.

Purpose of the Study:

• To address the limitations of arbitrary pseudocount usage in fold change calculations for high-throughput sequencing data.
• To develop a statistically rigorous method for selecting appropriate pseudocounts.
• To introduce a novel, robust estimator for fold changes that performs well with low counts.

Main Methods:

• Utilized a statistical framework for fold changes where pseudocounts represent parameters of a prior distribution for Bayesian inference.
• Proposed and evaluated an empirical Bayes procedure for selecting statistically rigorous pseudocounts.
• Introduced and tested the Ψ LFC estimator for fold changes, comparing its performance against existing methods using simulations and real data.

Main Results:

• Demonstrated that arbitrary pseudocount choices can introduce bias in fold change estimations.
• The proposed empirical Bayes procedure offers a statistically sound alternative for pseudocount selection.
• The novel Ψ LFC estimator exhibited favorable properties, showing reduced deviations from the true values, particularly for low-count data, compared to current methods.

Conclusions:

• The study provides a statistically rigorous approach to handling pseudocounts in fold change calculations for RNA-seq and similar experiments.
• The Ψ LFC estimator offers improved accuracy, especially for biological entities with limited sequencing reads.
• The findings have significant implications for the interpretation of differential expression analysis in sequencing studies.