In-Silico Tool for Predicting, Scanning, and Designing Defensins

Abstract

Defensins are host defense peptides present in nearly all living species, which play a crucial role in innate immunity. These peptides provide protection to the host, either by killing microbes directly or indirectly by activating the immune system. In the era of antibiotic resistance, there is a need to develop a fast and accurate method for predicting defensins. In this study, a systematic attempt has been made to develop models for predicting defensins from available information on defensins. We created a dataset of defensins and non-defensins called the main dataset that contains 1,036 defensins and 1,035 AMPs (antimicrobial peptides, or non-defensins) to understand the difference between defensins and AMPs. Our analysis indicates that certain residues like Cys, Arg, and Tyr are more abundant in defensins in comparison to AMPs. We developed machine learning technique-based models on the main dataset using a wide range of peptide features. Our SVM (support vector machine)-based model discriminates defensins and AMPs with MCC of 0.88 and AUC of 0.98 on the validation set of the main dataset. In addition, we created an alternate dataset that consists of 1,036 defensins and 1,054 non-defensins obtained from Swiss-Prot. Models were also developed on the alternate dataset to predict defensins. Our SVM-based model achieved maximum MCC of 0.96 with AUC of 0.99 on the validation set of the alternate dataset. All models were trained, tested, and validated using standard protocols. Finally, we developed a web-based service “DefPred” to predict defensins, scan defensins in proteins, and design the best defensins from their analogs. The stand-alone software and web server of DefPred are available at https://webs.iiitd.edu.in/raghava/defpred.

Recent development of machine learning-based methods for the prediction of defensin family and subfamily

“…Until now, a number of computational approaches for in silico prediction of defensins and their family/subfamily have been developed (Karnik et al, 2009[ 29 ]; Kaur et al, 2021[ 30 ]; Ramya Kumari et al, 2012[ 45 ]; Seebah et al, 2007[ 47 ]; Zuo and Li, 2009[ 62 ]; Zuo et al, 2015[ 61 ]). The general machine learning framework used for the prediction of defensins involves four major steps as summarized in Figure 1 (Fig.…”

“…We reviewed all the datasets used for developing the existing methods (Karnik et al, 2009[ 29 ]; Kaur et al, 2021[ 30 ]; Zuo and Li, 2009[ 62 ]; Zuo et al, 2015[ 61 ], 2019[ 60 ]). The detailed information of these datasets are provided in Table 2 (Tab.…”

“…The detailed information of these datasets are provided in Table 2 (Tab. 2) (References in Table 2: Karnik et al, 2009[ 29 ]; Kaur et al, 2021[ 30 ]; Ramya Kumari et al, 2012[ 45 ]; Zuo and Li, 2009[ 62 ]; Zuo et al, 2015[ 61 ], 2019[ 60 ]). As seen in Table 2 (Tab.…”

“…1) , including, the preparation of training and independent test datasets, feature extraction, feature optimization, and model development and evaluation. Currently, there are six state-of-the-art computational approaches that have been developed for the in silico prediction of defensins, including, Karnik's method (Karnik et al, 2009[ 29 ]), ID_RAAA (Zuo and Li, 2009[ 62 ]), Defensinpred (Ramya Kumari et al, 2012[ 45 ]), iDPF-PseRAAAC (Zuo et al, 2015[ 61 ]), iDEF-PseRAAC (Zuo et al, 2019[ 60 ]) and DEFPRED (Kaur et al, 2021[ 30 ]) as summarized in Table 1 (Tab. 1) (References in Table 1: Karnik et al, 2009[ 29 ]; Kaur et al, 2021[ 30 ]; Ramya Kumari et al, 2012[ 45 ]; Zuo and Li, 2009[ 62 ]; Zuo et al, 2015[ 61 ], 2019[ 60 ]).…”

“…1) into two groups according to their predictive applications. The first group is comprised of computational methods developed for the in silico prediction of defensins, which make up two out of six existing methods (i.e., Karnik's method (Karnik et al, 2009[ 29 ]) and DEFPRED (Kaur et al, 2021[ 30 ])). The second group is focused on those computational methods which have been developed for the in silico prediction of the defensin peptide family and subfamily, and comprise of four out of the six existing methods (i.e., ID_RAAA (Zuo and Li, 2009[ 62 ]), Defensinpred (Ramya Kumari et al, 2012[ 45 ]), iDPF-PseRAAAC (Zuo et al, 2015[ 61 ]) and iDEF-PseRAAC (Zuo et al, 2019[ 60 ])).…”