Probability waves: Adaptive cluster-based correction by convolution of p-value series from mass univariate analysis

Area of Science:

• Neuroscience
• Biostatistics
• Data Analysis

Background:

• P-value correction methods face challenges in large-scale exploratory data analysis, risking increased Type II errors or reduced Type I errors.
• Mass univariate analysis generates extensive p-value data, necessitating robust correction techniques.
• Identifying true null hypothesis rejections amidst numerous tests is a significant statistical hurdle.

Purpose of the Study:

• To develop and validate a novel cluster-based p-value correction method for mass univariate analysis.
• To address the limitations of existing p-value correction methods in handling large exploratory datasets.
• To improve the accuracy of statistical inference in neuroimaging and other high-throughput data analyses.

Main Methods:

• A new method was developed that analyzes patterns in probability vectors from mass univariate analysis to correct p-values.
• The method involves convolving the Log10 of the p-vector with a Gaussian window, with window length determined by signal autocorrelation.
• Monte-Carlo simulations were employed to assess confidence intervals, compare corrected and uncorrected p-vectors, and empirically verify Type I error rates using simulated ADHD and control subject data.

Main Results:

• The novel p-value correction method demonstrated minimal differences from simulated data and maximal differences from raw p-vectors when the window length was optimized by autocorrelation.
• Monte-Carlo simulations showed a 2.78 ± 4.83% difference from the corrected p-vector, contrasting sharply with a 596 ± 5.00% difference from the raw p-vector (p=0.0003).
• The proposed method proved less conservative than False Discovery Rate (FDR) methods, which rejected nearly all significant p-values.

Conclusions:

• The developed cluster-based method is biologically and statistically suitable for correcting p-values in mass univariate analysis of electroencephalography (ERP) waves.
• The method utilizes adaptive parameters, making it flexible for different types of neurophysiological data.
• This approach offers a more reliable alternative to existing methods, particularly for complex datasets where identifying true effects is crucial.