ProtNLS is a nuclear localization signal (NLS) prediction tool. Built on protein pretrained models, it innovatively integrates multi-level feature extraction, a channel-sequence dual-attention mechanism, and a learnable attention-unit aggregation strategy to accurately identify potential NLS regions from protein sequences and perform NLS protein classification. On an independent test set, the tool achieves strong performance (AUC 0.9746, accuracy 0.9191), combining high predictive power and good interpretability. It provides efficient and reliable AI-assisted support for subcellular localization studies and nuclear transport mechanism analysis.
The page returns classification probabilities, residue-level attention scores (Attention Map), and candidate NLS segment information, helping users quickly screen mutation sites, truncation fragments, or tag-fusion strategies in experimental design.
1. Protein Sequences (up to 10 FASTA entries):
Parsed sequences: 0, total residues: 0
Tool Notes
- Results include: predicted label, probability, residue attention map, candidate regions, and their statistical features
- Input supports multiple FASTA records (up to 10), with each sequence length <= 1024 aa
Model Performance Metrics
Test metrics | auc=0.9746 acc=0.9191 threshold=0.5
precision recall f1-score support
0 0.9455 0.9300 0.9377 4701
1 0.8712 0.8983 0.8845 2477
accuracy 0.9191 7178
macro avg 0.9083 0.9141 0.9111 7178
weighted avg 0.9199 0.9191 0.9193 7178