SPOTONE: Hot Spots on Protein Complexes with Extremely Randomized Trees via Sequence-Only Features

• Main Author: Preto, A. J.
• Publication Date: 2020
• Other Authors: Moreira, Irina S.
• Format: Article
• Language: eng
• Source: Repositórios Científicos de Acesso Aberto de Portugal (RCAAP)
• Download full: https://hdl.handle.net/10316/101280; https://doi.org/10.3390/ijms21197281
• topic: big-data; hot-spots; machine learning; protein–protein complexes; structural biology; Amino Acid Sequence; Amino Acids; Binding Sites; Computational Biology; Databases, Protein; Datasets as Topic; Humans; Protein Binding; Protein Interaction Mapping; Proteins; Machine Learning

Protein Hot-Spots (HS) are experimentally determined amino acids, key to small ligand binding and tend to be structural landmarks on protein-protein interactions. As such, they were extensively approached by structure-based Machine Learning (ML) prediction methods. However, the availability of a much larger array of protein sequences in comparison to determined tree-dimensional structures indicates that a sequence-based HS predictor has the potential to be more useful for the scientific community. Herein, we present SPOTONE, a new ML predictor able to accurately classify protein HS via sequence-only features. This algorithm shows accuracy, AUROC, precision, recall and F1-score of 0.82, 0.83, 0.91, 0.82 and 0.85, respectively, on an independent testing set. The algorithm is deployed within a free-to-use webserver at http://moreiralab.com/resources/spotone, only requiring the user to submit a FASTA file with one or more protein sequences.