Is your feature request related to a problem? Please describe.
Near-identical protein sequences (e.g. from different species or splice variants) are invisible to exact duplicate detection but introduce the same bias in downstream machine learning and statistical analyses. For immunological epitope datasets, homologous sequences from related pathogens may represent genuine cross-reactivity, not redundancy.
Describe the solution you'd like
Add optional protein homology clustering using CD-HIT (cd-hit) or Biopython pairwise alignment as fallback:
- Cluster at configurable identity threshold (default: 90 % for proteins, equivalent to
cd-hit -c 0.9)
- Report number of clusters, singleton ratio, and mean cluster size
- Flag columns where > 50 % of sequences fall into clusters of size ≥ 5 as potentially redundant
Activate via --homology CLI flag, same as DNA/RNA. Surface in the Protein Statistics tab.
Describe alternatives you've considered
BLAST-based all-vs-all alignment. Too slow for routine profiling; CD-HIT's greedy incremental clustering is the standard approach for this use case.
Additional context
Protein homology at 90 % identity is a meaningful biological threshold (roughly species-level conservation for many protein families). The threshold should be user-configurable via --homology-threshold to allow stricter (50 %, family-level) or looser analysis.
Similar to issue #9 (DNA/RNA Homology)
Is your feature request related to a problem? Please describe.
Near-identical protein sequences (e.g. from different species or splice variants) are invisible to exact duplicate detection but introduce the same bias in downstream machine learning and statistical analyses. For immunological epitope datasets, homologous sequences from related pathogens may represent genuine cross-reactivity, not redundancy.
Describe the solution you'd like
Add optional protein homology clustering using CD-HIT (
cd-hit) orBiopythonpairwise alignment as fallback:cd-hit -c 0.9)Activate via
--homologyCLI flag, same as DNA/RNA. Surface in the Protein Statistics tab.Describe alternatives you've considered
BLAST-based all-vs-all alignment. Too slow for routine profiling; CD-HIT's greedy incremental clustering is the standard approach for this use case.
Additional context
Protein homology at 90 % identity is a meaningful biological threshold (roughly species-level conservation for many protein families). The threshold should be user-configurable via
--homology-thresholdto allow stricter (50 %, family-level) or looser analysis.Similar to issue #9 (DNA/RNA Homology)