Classify intron sequences as U12-type (minor spliceosome) or U2-type (major spliceosome). A calibrated 42-model RBF-SVM ensemble scores each intron's splice-site and branch-point motifs — background-corrected against the species' own intron pool — into a species-agnostic probability (P_motif), and a per-species adjudicator decides whether the genome carries a detectable U12-type population.
pip install intronIC# Classify introns (loads default model automatically)
intronIC -g genome.fa.gz -a annotation.gff3.gz -n species_name -p 8
# Extract sequences without classification
intronIC extract -g genome.fa.gz -a annotation.gff3.gz -n species_name -p 8
# Verify installation with bundled test data
intronIC test -p 4Most eukaryotic introns (~99.5%) are spliced by the major (U2-type) spliceosome; a small fraction (~0.5%) are spliced by the minor (U12-type) spliceosome. U12-type introns carry a conserved TCCTTAAC branch point motif and have either AT-AC (~25%) or GT-AG (~75%) terminal dinucleotides. intronIC scores those motifs and classifies each intron.
Given a genome FASTA and a GFF/GTF annotation, intronIC runs five stages:
- Extract — pull introns from the annotation's CDS/exon coordinates (isoform handling, canonical/non-canonical terminus filters, partial-CDS handling).
- Score motifs — score each intron's 5′ splice site, branch point, and 3′ splice site against position-weight matrices built from a U12-type reference set.
- Background-correct — score against a background model built from the species' own intron pool, so a high score reflects genuine U12-type signal rather than that genome's base composition. These are the raw motif features.
- Classify — feed the raw features to a calibrated 42-model RBF-SVM ensemble → a per-intron ensemble margin →
P_motif, a single species-agnostic probability that the motif is U12-type (Platt-calibrated; also reported 0–100 asadjusted_score). - Adjudicate — at the genome level, test whether the strong U12-type calls form a real population standing above the species' own U2-type score tail (
z_excess, plus a call-strength gate) →motif_category∈ {DETECTED,INCONCLUSIVE,NOT_DETECTED,UNASSESSABLE}. An intron is labelledtype_id = u12whenP_motif ≥ 0.5and the genome is notNOT_DETECTED.
See Technical Details in the wiki for the full algorithm.
Where the Minor Things Are — WtMTA v2 (www.introns.info/WtMTA), a database of minor (U12-type) introns generated with intronIC v3 across ~2,600 eukaryotic genomes (~350M introns, ~1M high-confidence U12-type). Interactive species tree, splice-site logos, and bulk downloads.
intronIC v3 keeps the PWM motif scoring of the Moyer et al. (2020) method but replaces the per-species z-normalization that made scores comparable across genomes. That normalization assumed every genome matched the human reference geometry, and failed in two ways: it under-called divergent bearers (e.g. Amborella, Oryza) and produced false positives in genomes that have lost the minor spliceosome. v3 instead scores background-corrected raw motif log-odds — six features, an ensemble of 42 SVMs trained on 97 species across 14 clades — and calibrates the output to P_motif, a species-agnostic probability. All per-species reasoning moves to one output-level adjudicator, which reports whether a genome carries a detectable U12-type population (motif_category), referenced to that genome's own U2-type background.
| v1 (Moyer 2020) | v3 (current) | |
|---|---|---|
| Features | 3 motif scores | 6 raw features (→9 with interactions) |
| Classifier | 1 human-trained SVM | 42-model ensemble, 97 species |
| Cross-species | per-species z-normalization | species-agnostic P_motif |
| Per-species logic | folded into normalization | output-level adjudicator |
For the full comparison, see Technical Details → Relationship to the published method (v1).
- Pretrained, cross-species model loaded automatically — no per-run training or normalization step.
P_motif(0–1) /adjusted_score(0–100) per intron, plus a genome-levelmotif_categorycall.- Streaming mode (default) roughly halves peak memory on large genomes (~5.3 GB for full human at
-p 5) and is bit-identical to in-memory. - Parallel scoring via
-p N, and comprehensive per-intron metadata (phase, transcript position, parent gene/transcript).
Full documentation lives in the intronIC Wiki:
- Quick Start — Installation, dependencies, resource usage
- Overview — Classification approach and scientific background
- Output Files — File formats and score interpretation
- Technical Details — Algorithm, features, species adjudication
- Usage Info — Complete CLI reference
- Example Usage — Common workflows
- Changelog — Release notes and version history
If you use intronIC in your research, please cite:
Moyer DC, Larue GE, Hershberger CE, Roy SW, Padgett RA. (2020) Comprehensive database and evolutionary dynamics of U12-type introns. Nucleic Acids Research 48(13):7066-7078. doi:10.1093/nar/gkaa464
- intronIC Wiki — Documentation
- GitHub Issues — Bug reports
- GitHub Discussions — Questions and ideas
See CONTRIBUTING.md for guidelines.
git clone https://github.com/glarue/intronIC.git
cd intronIC
make install # Set up development environment
make test # Run tests