iAMP-CA2L: a new CNN-BiLSTM-SVM classifier based on cellular automata image for identifying antimicrobial peptides and their functional types

Abstract

Predicting antimicrobial peptides (AMPs’) function is an important and difficult problem, particularly when AMPs have many multiplex functions, i.e. some AMPs simultaneously have two or three functional classes. By introducing the ‘CNN-BiLSTM-SVM classifier’ and ‘cellular automata image’, a new predictor, called iAMP-CA2L, has been developed that can be used to deal with the systems containing both monofunctional and multifunctional AMPs. iAMP-CA2L is a 2-level predictor. The 1st level is to identify whether a given query peptide is an AMP or a non-AMP, while the 2nd level is to predict if it belongs to one or more functional types. As demonstration, the jackknife cross-validation was performed with iAMP-CA2L on a benchmark dataset of AMPs classified into the following 10 functional classes: (1) antibacterial peptides, (2) antiviral peptides, (3) antifungal peptides, (4) antibiofilm peptides, (5) antiparasital peptides, (6) anti-HIV peptides, (7) anticancer (antitumor) peptides, (8) chemotactic peptides, (9) anti-MRSA peptides and (10) antiendotoxin peptides, where none of AMPs included has ≥90% pairwise sequence identity to any other in the same subset. Experiments show that iAMP-CA2L has greatly improved the prediction performance compared with the existing predictors. iAMP-CA2L is freely accessible to the public at the web site http://www.jci-bioinfo.cn/ iAMP-CA2L, and the predictor program has been uploaded to https://github.com/liujin66/iAMP-CA2L.

A review on antimicrobial peptides databases and the computational tools

“…Xiao et al. designed a two-level classifier to first classify peptide sequences as an AMP, and then sub-classify them into 10 functional AMP categories ( 129 ).…”

Recent Progress in the Discovery and Design of Antimicrobial Peptides Using Traditional Machine Learning and Deep Learning

“…AMP discovery from large-scale natural known peptide libraries is based on the antimicrobial activity prediction from traditional ML models in a screening manner. Traditional ML techniques, such as SVM [ 92 , 93 , 94 , 95 , 96 ], discriminant analysis (DA) [ 97 ], RF [ 98 , 99 , 100 , 101 ], kNN [ 95 , 102 , 103 ], and ensemble learning [ 104 , 105 , 106 , 107 , 108 ] have been applied to discover AMPs by classification. Among these methods, SVM non-linearly transforms the original input space into a higher-dimensional feature space by means of kernel functions [ 109 , 110 ].…”

“…Xiao et al [ 96 ] proposed iAMP-CA2L, a two-layer DL predictor with cellular automata images (CAI) [ 155 ] as input features by the method of tandem fusion of CNN, Bi-LSTM, and SVM, to first identify AMPs and then 10 functional classes (ABPs, AVPs, AFPs, anti-biofilm peptides, anti-parasite peptides, anti-HIV peptides, ACPs, chemotactic peptides, anti-MRSA peptides, and anti-endotoxin peptides).…”

PrMFTP: Multi-functional therapeutic peptides prediction based on multi-head self-attention mechanism and class weight optimization

“…As July 2021, we conducted a literature query on Google Scholar with the keyword 'therapeutic peptide' and obtained 22 kinds of therapeutic peptide sequence datasets. There are a total of 22 types of therapeutic peptides: AAP [33], anti-bacterial peptide (ABP) [33,34], ACP [12,34], anti-coronavirus peptide (ACVP) [9], anti-diabetic peptide (ADP) [35], anti-endotoxin peptide (AEP) [34], anti-fungal peptide (AFP) [24,34], anti-HIV peptide (AHIVP) [34], anti-hypertensive peptide (AHP) [5], anti-inflammatory peptide (AIP) [33,36], anti-MRSA peptide (AMRSAP) [34], anti-parasitic peptide (APP) [13,34], anti-tubercular peptide (ATP) [37], anti-viral peptide (AVP) [33,34], blood-brain barrier peptide (BBP) [7], biofilm-inhibitory peptide (BIP) [8,34], chemotactic peptide (CP) [34], cell-penetrating peptide (CPP) [33], dipeptidyl peptidase IV peptide (DPPIP) [38], quorum-sensing peptide (QSP) [33], surfacebinding peptide (SBP) [33] and THP [11].…”

“…(3) the peptides with their number less than 40 were removed. In addition, CP was abandoned since there are too few CP to be statistically significant [34]. After these processes, we combined theses therapeutic peptide data and assigned the peptides with multi-label functions.…”