PlifePred is a computational platform developed for predicting the half-life of natural and modified peptides in mammalian blood using machine learning and sequence/structure-based approaches.
The platform assists researchers in designing therapeutic peptides with improved stability and bioavailability.
The study developed prediction models using:
- Amino acid composition
- Dipeptide composition
- Binary profiles
- Atom composition
- PaDEL chemical descriptors
Machine learning methods were used to estimate peptide half-life directly from peptide sequences and structures.
Web Server:
http://webs.iiitd.edu.in/raghava/plifepred/
In silico approaches for predicting the half-life of natural and modified peptides in blood
- Deepika Mathur
- Sandeep Singh
- Ayesha Mehta
- Piyush Agrawal
- Gajendra P. S. Raghava
PLOS ONE
01 June 2018
https://doi.org/10.1371/journal.pone.0196829 https://doi.org/10.5281/zenodo.20142268
Peptide therapeutics are increasingly important because of:
- High specificity
- Low toxicity
- Better tissue penetration
- Lower side effects
However, one major challenge is:
- Short half-life in blood due to enzymatic degradation
The half-life of peptides determines:
- Stability
- Bioavailability
- Dosing frequency
- Therapeutic efficacy
Experimental determination of peptide half-life is:
- Expensive
- Time-consuming
- Labor-intensive
PlifePred was developed to provide a fast computational alternative.
The peptide data was extracted from:
- PEPlife database
Initial collection:
- 1392 peptide entries
After filtering:
- Peptides length between 5–50 residues
- Half-life between 20 seconds and 24 hours
Final datasets:
- 261 unique peptides
- Natural + modified residues
- 163 natural peptides
Source: :contentReference[oaicite:1]{index=1}
The following input features were used:
Represents frequency of amino acids in peptide sequences.
Captures:
- Amino acid frequency
- Local sequence order
Binary encoding of terminal residues:
- N-terminal residues
- C-terminal residues
Frequency of:
- C
- H
- O
- N
- S
- F
- Cl
- Br
Generated using:
- PaDEL software
More than 15,000 descriptors were initially calculated.
Feature selection was performed using:
- CfsSubsetEval
- BestFirst algorithm
The following machine learning approaches were implemented:
- Support Vector Machine (SVM)
- SMOreg
- Linear Regression
- Gaussian Processes
- IBk (Instance-based learning)
Software used:
- SVM_light
- WEKA
Model evaluation was performed using:
- Leave-One-Out Cross Validation (LOOCV)
Source: :contentReference[oaicite:3]{index=3}
The study observed that peptides with:
Were enriched in:
- Glutamic acid (Glu)
- Alanine (Ala)
- Isoleucine (Ile)
- Leucine (Leu)
Were enriched in:
- Tyrosine (Tyr)
- Phenylalanine (Phe)
- Glycine (Gly)
- Histidine (His)
Aromatic residues were associated with lower peptide stability.
Source: :contentReference[oaicite:4]{index=4}
| Feature | Correlation (R) |
|---|---|
| Amino Acid Composition | 0.643 |
| Feature | Correlation (R) |
|---|---|
| Dipeptide Composition | 0.640 |
| Feature | Correlation (R) |
|---|---|
| Atom Composition | 0.532 |
Source: :contentReference[oaicite:5]{index=5}
Using:
- 45 selected PaDEL descriptors
| Method | Correlation (R) |
|---|---|
| SMOreg | 0.743 |
Additional metrics:
- MAE: 1.369
- RMSE: 1.932
Source: :contentReference[oaicite:6]{index=6}
Using:
- 43 selected PaDEL descriptors
| Method | Correlation (R) |
|---|---|
| SVM | 0.692 |
Additional metrics:
- MAE: 1.564
- RMSE: 2.075
Source: :contentReference[oaicite:7]{index=7}
The workflow included:
- Dataset extraction from PEPlife
- Dataset filtering
- Feature extraction
- Machine learning model development
- LOOCV validation
- Web server implementation
The workflow diagram is shown on page 3 of the paper. :contentReference[oaicite:8]{index=8}
PlifePred contains two major modules:
- Analog Generation
- Batch Submission
- Protein Scan
- Draw Module
- File Upload Module
Allows prediction for:
- Modified residues
- Chemically modified peptides
- Non-natural amino acids
Users can:
- Predict peptide half-life
- Generate mutants
- Design stable analogs
- Analyze physicochemical properties
Web Server:
http://webs.iiitd.edu.in/raghava/plifepred/
Source: :contentReference[oaicite:9]{index=9}
- SVM_light
- WEKA
- PaDEL Descriptor
- PHP
- Perl
- HTML
- Machine Learning Algorithms
PlifePred can be used for:
- Therapeutic peptide design
- Peptide engineering
- Drug discovery
- Stability prediction
- Rational peptide modification
- Pharmacokinetics research
The study demonstrated that:
- Chemical descriptors significantly improve prediction accuracy
- Structure-based models outperform sequence-only models
- Peptide composition strongly influences stability
- Machine learning can effectively predict peptide half-life
PlifePred provides a useful platform for rational design of stable therapeutic peptides.
Email: raghava@iiitd.ac.in
Address:
Indraprastha Institute of Information Technology Delhi
Creative Commons Attribution License (CC BY)