PERCEPT: Replacing binary p-value thresholding with scaling for more nuanced identification of sample differences

Abstract

Key to a biologists' capacity to understand data is the ability to make meaningful conclusions about differences in experimental observations. Typically, data are noisy, and conventional methods rely on replicates to average out noise and enable univariate statistical tests to assign p-values. Yet thresholding p-values to determine significance is controversial and often misleading, especially for omics datasets with few replicates. This study introduces PERCEPT, an alternative that transforms data using an ad-hoc scaling factor derived from p-values. By applying this method, low confidence effects are suppressed compared to high confidence ones, enabling clearer patterns to emerge from nois..