A random forest model for predicting exosomal proteins using evolutionary information and motifs

Abstract

Identification of secretory proteins in body fluids is one of the key challenges in the development of non-invasive diagnostics. It has been shown in the part that a significant number of proteins are secreted by cells via exosomes called exosomal proteins. In this study, an attempt has been made to build a model that can predict exosomal proteins with high precision. All models are trained, tested, and evaluated on a non-redundant dataset comprising 2831 exosomal and 2831 non-exosomal proteins, where no two proteins have more than 40% similarity. Initially, the standard similarity-based method BLAST was used to predict exosomal proteins, which failed due to low-level similarity in the dataset. To overcome this challenge, machine learning based models have been developed using compositional features of proteins and achieved highest AUROC of 0.70. The performance of the ML-based models improved significantly to AUROC of 0.73 when evolutionary information in the form of PSSM profiles was used for building models. Our analysis indicates that exosomal proteins have wide range of motifs. In addition, it was observed that exosomal proteins contain different types of sequence-based motifs, which can be used for predicting exosomal proteins. Finally, a hybrid method has been developed that combines a motif-based approach and an ML-based model for predicting exosomal proteins, achieving a maximum AUROC 0.85 and MCC of 0.56 on an independent dataset. The hybrid model in this study performs better than the presently available methods when assessed on an independent dataset. A web server and a standalone software ExoProPred has been created for the scientific community to provide service, code, and data. (https://webs.iiitd.edu.in/raghava/exopropred/). KeypointsO_LIExosomal proteins or non-classical secretory proteins are secreted by via exosomes C_LIO_LIA method has been developed for predicting exosomal proteins C_LIO_LIModels have been trained, tested, and evaluated on non-redundant dataset C_LIO_LIWide range of sequence motifs have been discovered in exosomal proteins C_LIO_LIA web server and standalone software have been developed C_LI