Fish mitochondrial genomes are closed, circular, double-stranded DNA molecules of roughly 16 kilobases. They contain 13 protein-coding genes, 22 transfer RNA genes, two ribosomal RNA genes and a D-loop. Their maternal inheritance, high copy number, rapid evolution and lack of recombination have made them valuable for reconstructing phylogenies, identifying species, studying populations and estimating selection pressure. Complete mitochondrial genomes are also widely available in resources such as NCBI GenBank and MitoFish. Before sequences downloaded from those databases can be used for analyses such as relative synonymous codon usage, or RSCU, and nonsynonymous-to-synonymous substitution ratios, or Ka/Ks, they must pass through quality control and have their lengths standardized.

That preprocessing is not straightforward because vertebrate mitochondria use a genetic code that differs from the standard code in several consequential respects. Under NCBI Translation Table 2, TGA represents tryptophan, so a program that treats it as a termination signal can incorrectly remove the final coding nucleotide and amino-acid residue. TAA and TAG are recognized stop codons, as are AGA and AGG in this code. In addition, many vertebrate mitochondrial genes are annotated with incomplete terminal codons, including T–, TA-, A– and AG-. These partial sequences are completed through post-transcriptional polyadenylation, which produces a functional termination signal. Public annotations may represent the same biological feature inconsistently, sometimes retaining a partial codon and sometimes presenting a completed stop codon.

Database sequences can also contain other features that make them unsuitable for codon analysis. An in-frame internal stop codon may indicate an insertion of nuclear DNA into mitochondrial DNA, often called a NUMT, or an annotation or sequencing error. IUPAC ambiguity symbols such as N, R and Y make a sequence unsuitable for the intended codon analyses and are excluded by the tool. In addition, codon-based calculations require sequence lengths that are strict multiples of three. A sequence that fails this condition after other processing can introduce systematic errors into estimates of codon composition or evolutionary divergence. Researchers have addressed these issues through manual inspection or custom scripts, but such approaches can be time-consuming, inconsistent and difficult to reproduce when processing hundreds or thousands of records.

FishMitoCleaner, developed as a single-file Python program and released as version 1.0.0, is intended to fill a gap between tools designed to annotate mitochondrial genomes or extract genes and the need to clean sequences that have already been extracted. It requires Python 3.8 or later and is described as running on Windows, macOS and Linux. The program accepts a multi-sequence FASTA file and processes each record independently, so sequences need not be sorted or aligned first. It also offers a demonstration mode that runs a built-in dataset when no input file is supplied. Required dependencies include Biopython for sequence handling, Matplotlib for figure generation and Tabulate for formatted console output. The developers state that the tool requires no programming skills to operate and that dependencies are checked at runtime, with informative errors displayed if a package is missing or outdated.

The program applies a five-step sequential workflow, stopping the processing of an individual sequence when it encounters an exclusion criterion. First, a character check accepts only the bases A, T, C and G; records containing ambiguity codes or other non-standard characters are marked as skipped. Second, the sequence is scanned for in-frame internal stop codons. TAA, TAG, AGA and AGG are treated as stop codons, while TGA is excluded from that scan because it encodes tryptophan under the vertebrate mitochondrial code. Third, a terminal TGA is protected and retained rather than trimmed. Fourth, the program classifies the sequence ending: a complete stop codon is removed, an incomplete terminal pattern is repaired by deleting its partial bases, or a sequence with no recognized stop is retained with a warning. Finally, retained sequences are trimmed as needed so that their lengths are multiples of three.

Each processed record receives one of five final status codes. CLEANED indicates that a complete or incomplete stop codon was removed, while RETAINED indicates that a terminal TGA was preserved. NO_STOP marks a sequence whose ending did not match a recognized stop-codon pattern but was kept with a warning. SKIPPED denotes exclusion because of invalid characters or an internal stop codon, and MOD3_FAIL is reserved for the rare case in which length normalization cannot be completed. The tool writes acceptable sequences to a FASTA file, prints a record-by-record report to the terminal and generates both raster and vector summary figures. The raster output is a 600-DPI PNG, while the PDF is a vector version intended for journal submission. A built-in test suite contains 32 unit and integration tests covering valid and invalid characters, the four complete stop codons, the four incomplete terminal patterns, terminal TGA retention, internal stop detection and length normalization; all 32 passed under Python versions 3.8 through 3.12.

For the real-world validation, the researchers downloaded complete mitochondrial genomes for 22 fish species from GenBank, extracted the 13 annotated protein-coding genes from each genome and processed the resulting 286 sequences with default settings. The species represented seven orders: Pleuronectiformes, Salmoniformes, Cypriniformes, Perciformes, Anguilliformes, Siluriformes and Tetraodontiformes. Of the 279 sequences processed successfully, 169, or 59.1% of the full set, ended in a complete stop codon and were trimmed by three base pairs. Another 110, or 38.5%, ended in an incomplete stop-codon pattern and were cleaned after that pattern was recognized. No sequence in this dataset ended in TGA, and no record received a NO_STOP status. The validation therefore demonstrated the complete and incomplete stop-codon paths in a fish dataset, but it did not include an observed terminal-TGA case or an unrecognized termination.

Seven of the 286 sequences, accounting for 2.4% of the validation set, were marked as skipped because they contained in-frame internal stop codons. The authors described these records as potentially reflecting NUMT contamination or errors in source-database annotations, although the screening result alone does not establish the precise cause of each sequence. The high proportion of incomplete terminations was interpreted as consistent with the biology of vertebrate mitochondrial genes and with reports from other teleost datasets, supporting the inclusion of partial stop-codon handling as a standard step rather than an exceptional procedure. FishMitoCleaner differs from MitoAnnotator, MITOS and MitoGeneExtractor because it is designed to quality-control and normalize already annotated coding sequences rather than annotate assemblies or extract genes from sequencing libraries. The authors position it as a complementary step between database retrieval and downstream analysis.

To assess whether cleaning affected a downstream measure, the researchers calculated RSCU values for all 60 sense codons before and after processing under the vertebrate mitochondrial code, then calculated the mean absolute difference for each of the 13 genes. Six genes—ATP8, COX1, COX2, COX3, ND3 and ND6—had mean absolute differences of exactly zero because all 22 sequences in each category remained the same length after cleaning. Seven genes each lost one sequence to the internal-stop filter and had mean absolute differences from 0.0099 for ND5 to 0.0229 for ND4L. Welch’s independent-samples t-tests detected no significant before-and-after differences for any gene or codon, with all p-values above 0.05 and a minimum of 0.62. These results indicate no detected change in RSCU within the tested dataset, while not establishing equivalence across all possible mitochondrial datasets or downstream metrics.

FishMitoCleaner has several limitations. It implements only the vertebrate mitochondrial code and cannot be used for invertebrate mitochondrial sequences with different coding rules. It expects coding sequences to have already been extracted and does not identify gene boundaries or annotate raw mitochondrial genomes. Records with unrecognized terminations are retained with a warning but require manual inspection. Length normalization trims trailing nucleotides when a coding sequence is not a multiple of three, but that adjustment may not be biologically appropriate when natural insertions or deletions are present; affected sequences therefore warrant individual review. Validation covered 22 species and focused primarily on RSCU; broader testing with more species, known mitochondrial structural variation, Ka/Ks estimates and amino-acid composition was identified as future work. Although the code is shared across vertebrate classes, the reported validation was confined to fish. The study presents the tool as a practical and reproducible bridge between public mitochondrial sequence databases and codon-level analyses, with demonstrated support for consistent stop-codon handling, sequence exclusion and length normalization in the tested data.

Subject of Research: Automated quality control and normalization of fish mitochondrial protein-coding sequences using the vertebrate mitochondrial genetic code

Article Title: FishMitoCleaner: Automated Stop Codon Processing and Quality Control of Fish Mitochondrial Protein‐Coding Sequences

Article References: Yang, S., Xu, P., Wang, J., & Cai, X. (2026). FishMitoCleaner : Automated Stop Codon Processing and Quality Control of Fish Mitochondrial Protein‐Coding Sequences. Ecology and Evolution, 16(10), Article e74557. https://doi.org/10.1002/ece3.74557

Image Credits: AI Generated

DOI: 10.1002/ece3.74557

Keywords: FishMitoCleaner, fish mitogenomes, mitochondrial DNA, protein-coding sequences, genetic code, quality control, stop codons, codon usage, bioinformatics, NUMTs, open-source software, Automated