Calibrating tetranucleotide-frequency distances for metagenomic binning with right-skewed distribution models

Metagenomic binning is a pivotal step in reconstructing metagenome-assembled genomes (MAGs) from complex microbial communities, and it critically depends on reliable measures of similarity between contigs. In many workflows, tetranucleotide-frequency (TNF) distances are translated into probabilistic evidence of a shared genome of origin. Despite their central role, these distances are often modeled with convenient but poorly matched assumptions, even though they are intrinsically non-negative and frequently exhibit pronounced right-skewness-features that can distort tail behavior and weaken downstream thresholding decisions. In this work, we introduce a likelihood-based framework for characterizing intra- and inter-genomic TNF distance distributions with flexible right-skewed parametric models and for converting fitted distributions into calibrated distance-to-probability scores within a MaxBin-style scheme. Our approach provides a principled statistical basis for distributional assessment, probability calibration, and transparent operating-point selection, with the goal of improving robustness and interpretability in TNF-driven binning.

All codes related to the article are available through a public GitHub repository at https://github.com/omar-hajjaji/Calibrating-TNF-Distances-for-Metagenomic-Binning-with-Right-Skewed-Distribution-Models.