Bio Alignment Trimming

HMMcleaner: Per-Residue Cleaning

Run HMMcleaner before column trimming when cross-contamination or annotation errors are suspected; it masks suspect residues with X rather than removing columns. Apply only to alignments with >= 15 sequences (below that the per-residue pHMM has too little signal and over-flags divergent residues).

HmmCleaner.pl input.fasta
HmmCleaner.pl input.fasta --no-large-remove

Output is input_hmm.fasta with low-confidence residues masked. Used in OrthoMaM and Compositional Heterogeneity-aware phylogenomic pipelines where cross-contamination from sequencing or annotation errors is suspected.

Sample-size sensitivity: HMMcleaner builds a per-residue HMM from the OTHER sequences at each position, so on small alignments (<15 sequences) the HMM has insufficient signal and over-flags real divergent residues as contamination. Di Franco et al 2019 show two related effects: pHMMs built from few sequences carry less signal, lowering HMMcleaner sensitivity, and its false positives are driven mainly by evolutionary divergence -- specificity falls with gap frequency, evolutionary rate, and the fraction of ambiguously-aligned regions rather than by a fixed sequence count (global specificity ~94% at default settings). OrthoMaM v11 and PhyloHerb pipelines apply HMMcleaner only to alignments with >= 15 sequences. For smaller alignments, manual inspection or BLAST-based contamination screening is more reliable.

Gblocks: The Legacy Default

Use Gblocks only when matching a legacy pipeline; prefer ClipKIT or trimAl for new work. Default parameters are too aggressive -- always relax with -b1=50% -b2=85% -b3=10 -b4=5 -b5=h.

Gblocks input.fasta -t=p -b1=50% -b2=85% -b3=10 -b4=5 -b5=h

PhyIN: Phylogenetic Incompatibility Trimming

PhyIN (Maddison 2024 PeerJ) is a complementary trimmer that flags neighbouring columns whose split patterns are phylogenetically incompatible (i.e. cannot share a tree). It targets a different failure mode than gap-based or entropy-based trimmers: ClipKIT, trimAl, and BMGE all evaluate columns independently, so a region with consistent gap structure passes their filters even if its character patterns are pairwise tree-incompatible (alignment artefact). PhyIN catches that case.

phyin input.fasta -o trimmed.fasta -w 5 -t 0.5

Use PhyIN as a SECOND-PASS trimmer after ClipKIT/trimAl when alignment artefact is the suspected source of incongruence in a phylogenomic dataset; not a replacement for first-pass column filtering. PhyIN's incompatibility test is signal-direction agnostic, so it does not distinguish "the alignment is wrong here" from "true incongruent locus" (e.g. introgression, ILS); only gene-tree comparison after tree-building can disambiguate. Reference: PeerJ 2024 paper.

Decision Tree by Downstream Goal

What is the next step?
+- Phylogenetic ML tree (RAxML, IQ-TREE)
|  +- Concatenated supermatrix? -> ClipKIT kpic-smart-gap
|  +- Single gene? -> ClipKIT smart-gap or trimAl -automated1
|  +- Deep prokaryotic? -> BMGE -h 0.4 -g 0.2
|  +- Suspected cross-contamination? -> HMMcleaner first, then ClipKIT
|
+- Bayesian inference (MrBayes, BEAST)
|  +- Same as ML; trimming is acceptable but log column mapping
|
+- HMM profile (HMMER hmmbuild, HHsuite)
|  +- trimAl -gappyout (aggressive gap removal helps profile quality)
|
+- Selection analysis (PAML codeml, HyPhy)
|  +- DO NOT aggressively trim
|  +- Use TCS column masking (T-Coffee -evaluate) or GUIDANCE2
|  +- See alignment/multiple-alignment confidence assessment
|
+- Sequence logo / motif scan
   +- Light gap-only trim (trimAl -gt 0.5)
   +- Preserve all variable positions

TCS Column Masking for Selection Analysis

TCS (Chang et al 2014 MBE) scores each column on a 0-9 scale by consistency across a pairwise-alignment library. The recommended dN/dS workflow:

Goal: Mask unreliable columns before selection analysis to prevent alignment errors from inflating false-positive dN/dS calls.

Approach: Build a library-rich consistency score via M-Coffee, evaluate the alignment against that library to obtain per-column TCS scores, then keep columns at or above a chosen confidence threshold using seq_reformat.

# 1. Generate library-rich consistency scores via M-Coffee (combines libraries from MAFFT, MUSCLE, ClustalW, ProbCons)
t_coffee input.fasta -mode mcoffee -output fasta_aln -outfile aligned.fasta

# 2. Compute per-column TCS scores against the same library
t_coffee -infile aligned.fasta -evaluate -output score_ascii > tcs_scores.ascii

# 3. Filter columns at a chosen threshold via seq_reformat (5-9 = retain columns scoring 5 or higher)
t_coffee -other_pg seq_reformat -in aligned.fasta -struc_in tcs_scores.ascii \
    -struc_in_f number_aln -action +keep '[5-9]' -output fasta_aln > aligned_tcs5.fasta

Threshold guidance: TCS >= 5 retains columns confidently aligned by the majority of library methods; TCS >= 7 is a stricter operational convention on the seq_reformat 0-9 display scale. Chang et al 2014 do not prescribe an integer 5/7 cutoff -- on BAliBASE 3 and PREFAB 4 structural benchmarks they keep residues scoring above ~0.6 on the native 0-1 scale and drop columns scoring below 2, with TCS outperforming GUIDANCE and HoT. TCS-from-mcoffee gives substantially better column-confidence ranking than TCS-from-default-tcoffee because the library is more diverse. For the PAML branch-site test, Fletcher & Yang 2010 showed alignment errors inflate false positives dramatically (up to ~0.99 with ClustalW, ~0.13 even with codon-aware PRANK) and that removing gappy columns did not reduce them; confidence-based column masking (TCS, GUIDANCE2) is a common mitigation but was not evaluated in that study, since TCS postdates it (Chang et al 2014).

MACSE Frameshift Markers Need Post-Processing

MACSE encodes detected frameshifts as ! (within-codon insertion) and * (premature stop) in the nucleotide output. PAML codeml interprets ! as missing data; HyPhy BUSTED/MEME interpret ! as N but * triggers a parse error mid-sequence. Standard cleanup before downstream analysis:

Goal: Convert MACSE frameshift and stop markers into formats that PAML and HyPhy will parse correctly.

Approach: Replace ! with gaps for PAML, and use MACSE's own exportAlignment sub-program with -codonForInternalStop NNN for HyPhy-safe output.

# Convert frameshift markers to gaps; replace internal stops with NNN
sed -e 's/!/-/g' aligned_nt.fasta > aligned_paml.fasta
java -jar macse_v2.jar -prog exportAlignment -align aligned_nt.fasta \
    -codonForFinalStop --- -codonForInternalStop NNN \
    -out_NT aligned_hyphy.fasta -out_AA aligned_hyphy_aa.fasta

-codonForInternalStop NNN replaces internal stops with NNN (HyPhy-safe). Without this step, the dN/dS run silently produces results that do not correspond to the alignment shown.

Aggressiveness Cap

Compute the retained-column fraction and warn if it drops below 0.7. Trimming more than 20-30% of columns rapidly degrades ML tree accuracy on empirical data.

def trimming_fraction(input_alignment, trimmed_alignment):
    return trimmed_alignment.get_alignment_length() / input_alignment.get_alignment_length()

If retention drops below 0.7, switch to a less aggressive mode or accept the original alignment unfiltered.

Column Mapping for Reproducibility

When trimming for phylogenetic input, retain the column-index mapping so downstream site-specific analyses (selection per site, structure-mapped residues, etc.) can be back-traced:

trimal -in input.fasta -out trimmed.fasta -automated1 -colnumbering > kept_columns.txt

clipkit input.fasta -m kpic-smart-gap --log -o trimmed.fasta

These two flags emit different formats. trimal -colnumbering writes a comma-separated list of the original column indices that survived (e.g. 0, 1, 2, 4, ...) on stdout. clipkit --log writes a <output>.log file with one row per original column reporting position\tkeep_or_trim\tsite_classification\tgap_proportion. To compare runs across the two tools, parse each format separately and reconcile to a common "kept-column index list" before applying the index map for site-specific cross-reference.

Common Errors

Related Skills

• alignment/multiple-alignment - Generate the input MSA before trimming; confidence assessment (GUIDANCE2, TCS, MUSCLE5 ensemble) for selection-analysis input
• alignment/msa-parsing - Parse the trimmed alignment; sequence weighting; coordinate mapping
• alignment/msa-statistics - Compute conservation and gap statistics to inform mode selection
• alignment/structural-alignment - Trim structural MSAs the same way as sequence MSAs
• phylogenetics/modern-tree-inference - Build trees from trimmed alignments

References

• Steenwyk JL, Buida TJ, Li Y, Shen XX, Rokas A. 2020. ClipKIT: a multiple sequence alignment trimming software for accurate phylogenomic inference. PLOS Bio 18:e3001007.
• Capella-Gutierrez S, Silla-Martinez JM, Gabaldon T. 2009. trimAl: a tool for automated alignment trimming in large-scale phylogenetic analyses. Bioinf 25:1972-1973.
• Criscuolo A, Gribaldo S. 2010. BMGE: a new software for selection of phylogenetic informative regions from multiple sequence alignments. BMC Evol Biol 10:210.
• Maddison WP. 2024. PhyIN: trimming alignments by phylogenetic incompatibilities among neighbouring sites. PeerJ 12:e18504.
• Tan G, Muffato M, Ledergerber C, Herrero J, Goldman N, Gil M, Dessimoz C. 2015. Current methods for automated filtering of multiple sequence alignments frequently worsen single-gene phylogenetic inference. Syst Biol 64:778-791.
• Fletcher W, Yang Z. 2010. The effect of insertions, deletions, and alignment errors on the branch-site test of positive selection. MBE 27:2257-2267.
• Chang JM, Di Tommaso P, Notredame C. 2014. TCS: a new multiple sequence alignment reliability measure. MBE 31:1625-1637.