HemoPI-MOD2

Prediction of Hemolytic Potency (HC50) of Chemically Modified Peptides

📚 Reference

Rathore AS et al. HemoPI-MOD2: Prediction of Hemolytic Potency (HC50) of Chemically Modified Peptides. #Coming Soon#

📌 Introduction

HemoPI-MOD2 is developed to help researchers estimate the hemolytic activity of peptides before synthesis. It predicts HC50, the concentration causing 50% hemolysis, as a regression value rather than a bare hemolytic/non-hemolytic call, so analogues can be ranked against each other.

Modified peptides are first-class input: N-terminal acetylation, C-terminal amidation, head-to-tail cyclization, D-amino acids and non-natural residues are all expressed in MAP notation and used by the models.

Three models are offered, each taking a different view of the same peptide, plus a Design module that enumerates and scores every single-residue analogue of a lead.

🔗 Visit the web server for more information: HemoPI-MOD2 Web Server

📖 Please cite relevant content for complete details, including the algorithm behind the approach.

🧪 Quick Start

# 1. Clone or unpack the standalone
cd hemopimod2

# 2. Set up the environment (conda recommended)
conda env create -f environment.yml
conda activate HemoPI-MOD2

# 3. See the available options
python HemoPI_MOD2.py -h

# 4. Run prediction on the sample input
python HemoPI_MOD2.py -i example_sequence.fasta -o output.csv -j 1 -m 1

# 5. Design all single-residue analogues of a lead
python HemoPI_MOD2.py -i example_design.fasta -o design.csv -j 2 -m 1

The three model files are already in models/; nothing needs downloading.

🛠️ Installation Options

🧰 Pip Installation

To install HemoPI-MOD2 via PIP, run:

pip install hemopimod2

To check available options, type:

hemopimod2 -h

🔹 Standalone Installation

HemoPI-MOD2 is written in Python 3 and requires the following dependencies:

✅ Required Libraries

python>=3.9
pip install scikit-learn==1.6.1
pip install xgboost==3.2.0
pip install "numpy>=1.23,<2.0"
pip install pandas
pip install joblib

Needed only for the Chemical (-m 2) and Structure (-m 3) models:

pip install rdkit
pip install mordred==1.2.0
pip install "networkx>=2.8,<3.0"

Or install everything at once:

pip install -r requirements.txt

🔹 Installation using environment.yml

1. Create a new Conda environment:

conda env create -f environment.yml

2. Activate the environment:

conda activate HemoPI-MOD2

⚠️ Important Note

• scikit-learn and xgboost are pinned on purpose. The three model files in models/ were fitted with scikit-learn 1.6.1, and chemical_model.pkl holds a serialized XGBoost regressor. Other versions usually load, but are not guaranteed to reproduce the published values.
• The Sequence (-m 1) and Design (-j 2) modules run without RDKit and Mordred. Install those two only if you need the Chemical or Structure model.

🔬 Prediction

HemoPI-MOD2 regresses pHC50 and reports HC50 in µM, where HC50 = 10^(−pHC50). A higher HC50 means a less hemolytic peptide. By convention, HC50 above 100 µM is read as non-hemolytic. This is the cut-off published with the method and is fixed.

🔹 Model Options

• Sequence (MAP): amino acid composition + modification tags + D/L counts
• Chemical (SMILES): Mordred 2D descriptors
• Structure (PDB): Mordred 2D descriptors from the 3D structure

🚀 Usage

🔹 Minimum Usage

python HemoPI_MOD2.py -h

To run an example:

python HemoPI_MOD2.py -i example_sequence.fasta

🔹 Full Usage

usage: HemoPI_MOD2.py [-h]
                      -i INPUT
                      [-o OUTPUT]
                      [-j {1,2}]
                      [-m {1,2,3}]
                      [-d {1,2}]
                      [-wd WORKING DIRECTORY]

Arguments

Job type 1 (Prediction) runs with any of the three models. Job type 2 (Design) runs on the Sequence (MAP) model only.

-d filters only what is printed to the screen. The output CSV always holds every input peptide.

📂 Input & Output Files

✅ Input File Format

HemoPI-MOD2 supports two formats for peptides:

FASTA format (example: example_sequence.fasta)

Modifications are written in MAP notation, as plain one-letter amino acid codes with modifications in braces:

>example
{nt:Acet}KWKLFKKIL{ct:Amid}

For the Structure model, -i takes either one .pdb file or a folder of .pdb files; each structure's ID is taken from its file name.

✅ Output File

• Results are saved in CSV format.
• If no output file is specified, results are stored in output.csv.

Prediction (-j 1)

Under the Chemical model, an entry supplied as a peptide sequence is reported with both its sequence and its SMILES; an entry supplied as SMILES has no sequence form, so only its SMILES is shown. A long SMILES is shortened on the screen to keep the table readable, and written in full to the CSV.

Design (-j 2)

🔍 Jobs & Features

🔹 Job Types

🔹 The Design module

Every position of each input peptide is substituted by each of the other 19 amino acids, and every variant is scored with the Sequence model.

Modifications are never altered. Terminal tags ({nt:Acet}, {ct:Amid}, {cyc:N-C}), residue-level tags ({d}) and non-natural residues ({nnr:...}) are carried into every variant exactly as supplied; only the single-letter residue code at the scanned position changes. A non-natural residue occupies a chain position but is not itself mutated.

More than one lead can be submitted at once, and Fold_Change_vs_WT is measured against each lead's own wild type.

Limits: 50 residues per peptide (50 × 19 = 950 variants), and the Sequence model only. A per-variant Chemical scan would need a SMILES and descriptor pass for every mutant, and a Structure scan a folded PDB for every mutant; neither is practical.

🧠 The Models

📑 Package Contents

🚀 Start predicting and designing hemolytic peptides with HemoPI-MOD2 today!

🔗 Visit: HemoPI-MOD2 Web Server

(C) 2026 Dr. G.P.S. Raghava