Alignment Trimming - Usage Guide

Overview

This skill covers post-alignment column filtering and per-residue cleaning of multiple sequence alignments. It guides tool selection (ClipKIT, trimAl, BMGE, Divvier, HMMcleaner, Gblocks) by downstream goal (phylogenetic-tree input, HMM profile building, selection analysis, structure modelling) rather than treating "trimming" as a single decision.

Prerequisites

# CLI tools (all primary trimmers)
conda install -c bioconda clipkit trimal
# OR pip:
pip install clipkit
# trimAl: https://github.com/inab/trimal

# BMGE (Java)
# https://gitlab.pasteur.fr/GIPhy/BMGE

# Divvier (compile from source)
# https://github.com/simonwhelan/Divvier

# HMMcleaner (Perl + HMMER required)
# cpanm Bio::MUST::Apps::HmmCleaner

Quick Start

Tell your AI agent what you want to do: - "Trim this alignment for phylogenetic-tree input using ClipKIT" - "Prepare an HMM-ready alignment with aggressive gap removal" - "Remove cross-contaminating residues with HMMcleaner before phylogenomic analysis" - "Generate a structural-MSA-friendly trimming that preserves codon boundaries"

Example Prompts

Phylogenetic Input Trimming

"I have a concatenated supermatrix of 500 genes; what trimmer should I use for IQ-TREE input?"

"Trim this single-gene alignment with ClipKIT smart-gap and report what fraction of columns were retained"

"Compare ClipKIT kpic-smart-gap vs trimAl -strictplus on the same alignment and tell me which produces a more stable ML tree"

HMM Profile Preparation

"Trim this alignment for HMMER hmmbuild input using trimAl gappyout mode"

"Strip insert columns and lowercase characters before building a profile HMM"

Cross-Contamination Cleaning

"I suspect some sequences have contaminating residues; clean them with HMMcleaner first then trim"

"Mask low-confidence residues with HMMcleaner before building a tree"

Selection-Analysis-Safe Filtering

"Mark unreliable columns for codeml input but DO NOT remove them; selection analysis is sensitive to column removal"

"Use TCS column scores to mask unreliable codon positions before BUSTED/MEME/aBSREL"

Mode Comparison

"Run all three ClipKIT modes (smart-gap, kpic-smart-gap, gappy) and compare retention fractions"

"Test how aggressive each trimAl mode is on my alignment and recommend one"

What the Agent Will Do

Identify the downstream goal (tree, HMM, selection analysis, motif scan, structure modelling)
Select the appropriate trimmer and mode (ClipKIT modes vs trimAl modes vs BMGE entropy threshold)
Run with reproducibility-preserving flags (column mapping, log file)
Compute trimming fraction and warn if > 20-30% of columns are removed
For selection analysis, prefer per-column reliability scores (TCS, GUIDANCE2) over column removal
Optionally compare alternative trimming modes on the same input

Key Concepts

Concept	Meaning
Column retention	Fraction of original columns kept; > 70% is usually safe, < 50% concerning
Aggressiveness	How readily a mode removes columns; ClipKIT smart-gap is moderate, trimAl strictplus is high
Column mapping	Index list of which original columns survived (`-colnumbering`, `--log`)
Per-residue masking	Replace bad residues with `X` instead of removing the column (HMMcleaner)
Column splitting	Divvier alternative: split ambiguous columns rather than remove them
Codon-aware trimming	Preserve frame boundaries for selection analysis (MACSE trimAlignment)

Tips

ClipKIT kpic-smart-gap is the recommended default for phylogenetic-tree input on concatenated supermatrices; smart-gap for single genes
trimAl -gappyout is the right tool for HMM profile preparation; aggressive gap removal benefits HMMER hmmbuild
For selection analysis (PAML codeml, HyPhy), DO NOT aggressively trim; column removal causes false-positive selection signals (Fletcher & Yang 2010 MBE)
BMGE is the prokaryotic phylogenomics standard (default in GToTree); recommended -h 0.4 -g 0.2
Divvier preserves more phylogenetic signal than column removal because ambiguous columns are SPLIT rather than dropped
HMMcleaner masks contaminating residues with X; pair with a column trimmer for double cleaning
Always retain column-mapping (-colnumbering, --log) for reproducibility
The 20%/40% rule reconciles the Tan 2015 vs Steenwyk 2020 controversy: <20% column removal is neutral; >40% removes phylogenetic signal alongside noise. If your trimmer drops more than 40% of columns, the mode is too aggressive for the dataset
For phylogenomic incongruence due to suspected alignment artefact (rather than ILS or introgression), apply PhyIN (Maddison 2024 PeerJ) as a second-pass trimmer after ClipKIT/trimAl; it flags neighbouring columns whose split patterns are pairwise tree-incompatible
Gblocks default parameters are too aggressive for modern use; relax them or switch to ClipKIT/trimAl
Codon-aware trimming (MACSE trimAlignment or per-codon-block ClipKIT) is required for any post-MSA cleaning of dN/dS input

alignment/multiple-alignment - Generate the input MSA; confidence assessment (GUIDANCE2, TCS) for selection input
alignment/msa-parsing - Parse and post-process trimmed alignments
alignment/msa-statistics - Conservation and gap statistics to inform trimmer choice
alignment/structural-alignment - Apply same trimming logic to structural MSAs
phylogenetics/modern-tree-inference - Build trees from trimmed alignments