💻 bioinformatics

WitChi: Efficient Detection and Pruning of Compositional Bias in Phylogenomic Alignments Using Empirical Chi-Squared Testing

WitChi is a computationally efficient tool that uses empirical chi-squared testing to detect and iteratively prune compositionally biased sites from large-scale phylogenomic alignments, thereby restoring accurate phylogenetic topologies without the high computational cost of complex composition-aware models.

Original authors: Koestlbacher, S., Panagiotou, K., Tamarit, D., Ettema, T.

Published 2026-06-13

📖 3 min read☕ Coffee break read

CC BY 4.0

Original authors: Koestlbacher, S., Panagiotou, K., Tamarit, D., Ettema, T.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine you are trying to build a family tree for a huge group of people, but instead of using their actual family history, you're trying to guess their relationships based on what they wear. If a bunch of unrelated people all happen to wear the same bright red hat because it's trendy, your family tree might mistakenly group them together as close relatives, even though they aren't.

In the world of biology, this "bright red hat" is called compositional bias. It happens when unrelated species independently evolve to have similar chemical makeups in their DNA or proteins. This tricks scientists into drawing the wrong evolutionary family tree, especially when looking at very ancient or diverse groups of life.

Usually, scientists have to use super-complex, heavy-duty math to fix these mistakes, which is like trying to solve a puzzle with a giant, slow-moving robot. It takes too long for huge datasets.

Enter WitChi, a new tool that acts like a smart, fast detective for these biological puzzles. Here is how it works, using simple analogies:

The Detective Work: WitChi scans through the "pages" of a biological alignment (the list of DNA or protein sequences) and asks, "Does this specific column of data look suspiciously different from what we'd expect by chance?" It uses a statistical test called the "Chi-Squared test" to measure how weird the data looks.
The "What If" Game: To know if something is truly weird, WitChi plays a game of "what if." It shuffles the data around in a way that keeps the family tree structure intact but randomizes the specific details. If the real data still looks way more strange than the shuffled "what if" versions, it knows it's found a bias.
The Cleanup Crew: Once it spots the troublemakers (the columns causing the confusion), WitChi doesn't just delete the whole page. Instead, it uses three different strategies to carefully snip out only the most confusing parts, one by one, until the data looks "normal" again.
The Report Card: It gives scientists a clear score (Z-scores and p-values) to say exactly how bad the bias was and how much they cleaned it up.

Why is this a big deal?
The paper shows that WitChi is incredibly fast and efficient. While other tools struggle or take forever with massive datasets, WitChi scales up easily. The authors tested it on a massive dataset of 5,869 archaeal species (a type of single-celled organism) with over 10,000 sites of data.

Speed: WitChi finished the job in under one hour using just four computer cores.
Accuracy: After WitChi cleaned the data, the resulting family tree correctly identified key groups of organisms that were previously only visible when using much slower, more complex methods.

In short, WitChi is a fast, lightweight tool that sweeps away the "noise" of misleading chemical similarities, allowing scientists to see the true evolutionary family tree clearly, even when dealing with thousands of species.

Technical Summary: WitChi – Efficient Detection and Pruning of Compositional Bias

Problem Statement
Convergent evolution frequently leads to unrelated taxa independently evolving similar nucleotide or amino acid compositions. This phenomenon introduces compositional bias into biological sequence data, which significantly distorts phylogenetic inference, particularly within deep or unevenly sampled phylogenomic datasets. While composition-aware models exist to mitigate these distortions, their high computational demands often render them impractical for large-scale analyses involving thousands of taxa. Consequently, there is a critical need for efficient methods to detect and remove such biases without relying on computationally intensive evolutionary models.

Methodology
The authors present WitChi, a computationally efficient tool designed to identify and prune compositionally biased alignment columns using empirical significance testing. The methodology operates through the following core mechanisms:

Empirical Chi-Squared Testing: WitChi calculates taxon-specific chi-squared ( $\chi^2$ ) scores for alignment columns.
Null Distribution Generation: To establish statistical significance, the tool compares observed scores against null distributions derived from permutations within alignment columns. Crucially, these permutations are constrained to preserve the phylogenetic structure of the alignment, ensuring that the null model respects the underlying evolutionary relationships.
Iterative Pruning: The algorithm identifies sites most responsible for deviations from the expected null distribution. These sites are iteratively pruned using one of three available scoring algorithms. The process continues until the compositional bias is no longer statistically detectable.
Statistical Output: The tool generates interpretable metrics, including Z-scores and p-values, for both individual taxa and the entire alignment, quantifying the magnitude of the detected bias.

Key Results
The paper validates WitChi through both simulated and real-world datasets:

Simulation Studies: Pruning of simulated alignments with compositional heterogeneity demonstrated that WitChi reliably restores correct phylogenetic topologies when analyzed under standard, compositionally stationary models.
Benchmarking: In comparative benchmarks, WitChi outperformed BMGE (a widely used stationary-based trimming tool). Furthermore, the tool exhibits linear scaling with respect to the number of taxa, addressing the scalability limitations of previous methods.
Large-Scale Application: When applied to the archaeal GTDB r220 dataset (comprising 5,869 taxa and 10,101 sites), WitChi completed the pruning process in under one hour using four CPU cores. The resulting phylogeny successfully recovered key clades that had previously required in-depth analyses using complex models of sequence evolution to resolve.

Significance and Claims
The paper positions WitChi as an efficient and scalable solution for detecting and removing compositional bias in phylogenomic datasets ranging from thousands to tens of thousands of taxa. By enabling the use of standard, computationally tractable models on large datasets while mitigating the distorting effects of compositional heterogeneity, WitChi facilitates more accurate phylogenetic inference across the tree of life. The authors emphasize that the tool bridges the gap between the necessity of composition-aware analysis and the practical constraints of computational resources in large-scale phylogenomics.

Technical Summary: WitChi – Efficient Detection and Pruning of Compositional Bias

More like this