Abstract
The organization of homologous protein sequences into multiple sequence alignments (MSAs) is a cornerstone of modern analysis of proteins. Recent focus on the importance of alternatively-spliced isoforms in disease and cell biology has highlighted the need for MSA software that can appropriately account for isoforms and the exon-length insertions or deletions that isoforms may have relative to each other. We previously developed Mirage, a software package for generating MSAs for isoforms spanning multiple species. Here, we present Mirage2, which retains the fundamental algorithms of the original Mirage implementation while providing substantially improved translated mapping and improving several aspects of usability. We demonstrate that Mirage2 is highly effective at mapping proteins to their encoding exons, and that these protein-genome mappings lead to extremely accurate intron-aware alignments. Additionally, Mirage2 implements a number of engineering improvements that simplify installation and use.
Introduction
In bioinformatics, Multiple Sequence Alignment (MSA) is a fundamental task. However, standard MSA tools (like MAFFT or Muscle) were largely designed to handle sequences that differ by small mutations (substitutions, insertions, or deletions). They struggle with alternative splicing, a process where a single gene can produce multiple protein "isoforms" by including or excluding entire chunks of genetic code called exons.
When a standard tool sees a large gap in one sequence compared to another, it often tries to "force" an alignment to maximize a similarity score, leading to "over-alignment" where amino acids from completely different genomic regions are incorrectly paired. Mirage2 addresses this by using the genome as a ground-truth coordinate system.
Methods
The core philosophy of Mirage2 is "map first, align second." The pipeline follows a cascading logic to ensure high-quality results.
- Protein-to-Genome Mapping: Instead of comparing proteins directly, Mirage2 maps each protein sequence back to its encoding exons on the genome. It uses a hierarchy of methods: FastMap (a new native tool using GTF annotations), Spaln2 (for spliced-alignment), and BLAT (as a last resort for whole-genome searches).
- Intra-species Alignment: For isoforms within the same species, the tool uses genomic coordinates to create an MSA. Since these proteins originate from the same DNA, shared coding regions should theoretically have 100% identity.
- Inter-species Alignment: To align across species (e.g., human vs. mouse), Mirage2 performs a profile-to-profile alignment using a "splice-aware" scoring system that penalizes gaps differently if they occur at known splice sites.
The system was evaluated using 73,573 protein sequences from UniProtKB/SwissProt across human (GRCh38), mouse (GRCm38), and rat (RGSC 6.0) reference genomes.
Key Results and Insights
Mirage2 successfully mapped 99.12% of the SwissProt dataset to the genome. A major quality metric used is the "100% identity benchmark": within a single species, an optimal isoform alignment must reflect the exonic structure. If two isoforms share a coding region, the amino acids translated from that region should be identical.
In human intra-species alignments, general tools like Clustal-Omega showed a ≥10% reduction in column identity for 11% of alignments compared to Mirage2. Furthermore, Mirage2 avoids "length compaction," a phenomenon where general-purpose tools produce shorter alignments by "squishing" unrelated sequences together to avoid gap penalties.
Efficiency is also a priority. While the full pipeline took 265 minutes, disabling the whole-genome BLAT search (using the --blat_off flag) dropped the runtime to just 50.5 minutes—50% faster than its closest competitor, MAFFT, while still maintaining a 93.24% mapping rate.
Discussion and Conclusion
Mirage2 provides a robust solution for isoform-aware MSA, though it relies on high-quality reference genomes and annotations. While Spaln2 can occasionally introduce errors with "micro-exons" (shorter than 5 amino acids), Mirage2 includes post-processing to mitigate these issues.
Future applications of these high-quality protein-to-genome mappings include exploring uncommon splicing activity and identifying cross-species post-translational modification targets. Engineering improvements, such as a Docker image and automated scripts for downloading genomes, significantly lower the barrier to entry for researchers.
Summary updated