SpotON

Determination of Hot-Spots at protein-protein interfaces

SpotON is an ensemble machine learning web server for the high-accuracy identification and classification of interfacial residues in protein–protein complexes as Hot-Spots (HS) or Null-Spots (NS).

Trained on 53 non-redundant complexes using over 880 computed features, spanning 3D structure, sequence conservation, and physicochemical properties, SpotON applies a logistic regression ensemble of the best-performing classifiers across five hierarchical clusters.

With an F1-score of 0.91, it represents the highest accuracy reported in the literature for hot-spot prediction, offering a freely accessible, no-programming-required alternative to costly experimental alanine scanning mutagenesis.

Available Features

• PDB structure upload: Submit a protein complex PDB file directly, no preprocessing or scripting required
• Rapid web-based prediction: Results delivered via email upon cluster job completion, no local installation required
• Ensemble ML classifier: Combines Random Forest, SVM (polynomial) and PDA models via logistic regression meta-learning
• Per-residue HS/NS classification: Each interfacial residue is scored and classified with associated confidence for downstream use
• Balanced class training: Minority-class up-sampling prevents bias toward null-spots, ensuring high sensitivity
• State-of-the-art benchmark: Outperforms all previously published HS prediction methods on the independent test set

---

Identify binding hot-spots in your protein complex: upload a PDB file and receive per-residue HS/NS classifications. No programming knowledge required.

---