PPTPP: a novel therapeutic peptide prediction method using physicochemical property encoding and adaptive feature representation learning

• Published in: Bioinformatics Vol. 36; no. 13; pp. 3982 - 3987
• Main Authors: Zhang, Yu P, Zou, Quan
• Format: Journal Article
• Language: English
• Published: England Oxford University Press 01-07-2020
• Subjects: Machine Learning; Peptides
• ISSN: 1367-4803; 1460-2059; 1367-4811
• DOI: 10.1093/bioinformatics/btaa275

Abstract
Motivation
Peptide is a promising candidate for therapeutic and diagnostic development due to its great physiological versatility and structural simplicity. Thus, identifying therapeutic peptides and investigating their properties are fundamentally important. As an inexpensive and fast approach, machine learning-based predictors have shown their strength in therapeutic peptide identification due to excellences in massive data processing. To date, no reported therapeutic peptide predictor can perform high-quality generic prediction and informative physicochemical properties (IPPs) identification simultaneously.
Results
In this work, Physicochemical Property-based Therapeutic Peptide Predictor (PPTPP), a Random Forest-based prediction method was presented to address this issue. A novel feature encoding and learning scheme were initiated to produce and rank physicochemical property-related features. Besides being capable of predicting multiple therapeutics peptides with high comparability to established predictors, the presented method is also able to identify peptides’ informative IPP. Results presented in this work not only illustrated the soundness of its working capacity but also demonstrated its potential for investigating other therapeutic peptides.
Availability and implementation
https://github.com/YPZ858/PPTPP.
Supplementary information
Supplementary data are available at Bioinformatics online.