ProRafts A machine-learning predictor for raftophilicity, the protein affinity for biomembrane rafts.

BackgroundProtein raftophilicity refers to the affinity of proteins for cell biomembrane lipid domains, called rafts. Rafts are fluctuating nanoscale platforms that are enriched in cholesterol and sphingolipids, and that are considered relevant for cell signalling, viral function, and biomembrane trafficking. The dynamic partitioning of proteins into rafts depends on the physical and physico-chemical properties of the biomembranes where such proteins are embedded or attached; however it also depends on specific protein "features", such as acylation, glypidation, specific amino acid sequence motifs, transmembrane hydrophobic length, and surface accessible area to solvent. In this paper we present a method, and the resulting "ProRafts" predictor, that can be used to predict if a given mammal protein may be "raftophilic" or "non-raftophilic", without having an a priori knowledge of the physical and physico-chemical properties of the biomembranes where such protein is embedded or attached. ProRafts is based on a machine-learning algorithm, XGBoost, where data regarding the features of known raftophilic human-proteins fed the algorithm. ResultsProRafts enabled to predict correctly more than 80% of human proteins that are a priori known to be raftophilic; this is a promising result considering the limited size of the training dataset that we could build with data retrieved from protein databases. In addition, although we used protein features of known human raftophilic proteins, it was possible to identify accurately raft-proteins from other mammals than humans, such as mouse and rats. This finding suggests that certain protein features are sufficient to predict raftophilicity of proteins from different species. Moreover, our results indicated that phosphorylation may play a more relevant role for protein raftophilicity than indicated by previous studies. ConclusionRaftophilic proteins can be used as biomarkers in medical research, or can serve as targeting sites for therapeutics. In this respect, the machine learning method presented in this paper is a useful tool to guide experimental validations of raftophilicity of proteins in biomembranes, and facilitate the choice of proteins that can be used for experiments on biomimetic membranes.