NEPTUNE #

Introduction #

NEPTUNE is support vector machine-based classifier for tumor homing peptides (THPs) from the probabilistic information generated by the optimal baseline models. THPs are short peptides (i.e, 3-30 Amino acids). Its used for tumor diagnosis and therapeutic applications, such as tumor site drug-delivery. NEPTUNE is stacked ensemble learning approach, which able to secure THPs prediction accuracy while compare to existing methods (i.e., THpred, SCMTHP, and MIMML). Its two-layer prediction model, which in layer one, its produce the probabilities from different ensemble models called baseline models and those subjected to meta-modeler with stacking strategy to enhance the prediction accuracy.

Dataset: #

Main training set : 490 THPs and 490 Non-THPs Small training set : 350 THPs and 350 Non-THPs Main Independent dataset : 161 THPs and 161 non-THPs Small Independent dataset : 119 THPs and 119 non-THPs

Features: #

1. Amino Acid Composition (AAC)
2. Di-peptide Composition (DPC)
3. Amino Acid Index (AAI)
4. Amphiphilic pseudo-amino acid composition (APAAC)
5. Composition transition and distribution (CTD)
6. Pseudo-amino acid composition (PAAC)
7. Physicochemical properties (PCP)
8. Reduced protein sequences (RSs) based on acidity (RSacid)
9. Charge (RScharge)
10. DHP (RSDHP)
11. Polarity (RSpolar)
12. Secondary structure (RSsecond)

Baseline Machines: #

1. Random Forest (RF)
2. Support Vector Machine (SVM)
3. Partial Least Squares (PLS)
4. Logistics regression (LR)
5. Extremely randomized trees (ET)
6. K-nearest neighbor (k-NN)

Metamodel Machine: #

1. Support vector machine (SVM)

Features Importance #

1. Shapley additive extension (SHAP)

Webserver: #

1. http://pmlabstack.pythonanywhere.com/NEPTUNE

Reference : #

1. Charoenkwan, P., et al., NEPTUNE: A novel computational approach for accurate and large-scale identification of tumor homing peptides. Computers in Biology and Medicine, 2022: p. 105700.