ACP-designer

ML-powered toolbox for rapid design of Mastoparan-like anticancer peptides

Developed by Robert Vacha group • CEITEC

ACP-designer is a machine learning-based toolkit designed to accelerate the development of novel peptide-based anticancer therapeutics.

The toolbox combines:

• lightweight generative model
• cytotoxicity prediction filter
• physicochemical feasibility checks

→ allowing fast generation and filtering of promising Mastoparan-like anticancer peptide candidates.

---

✨ Key Features

1. Generative Model

Small character-level RNN trained on known anticancer peptides and fine-tuned specifically on Mastoparan sequences.

2. Cytotoxicity Filter

Classifier based on ESM-2 protein language model embeddings + Random Forest regressor
→ predicts probability of cytotoxicity. This probability is converted into three classes:

Output labels:

• LOW
• MEDIUM
• HIGH

3. Additional Quality Filters

Comparison of generated sequences against training set distribution in:

• net charge
• hydrophobic moment

Labels:

• OK – good agreement with known sequences
• EDGE – significant deviation (may cause synthesis/solubility problems)

---

Installation

Local Installation

1. Clone the Repository

   git clone https://github.com/karatedava/ACP-designer.git
   cd ACP-designer

2. Install Miniconda
   Download and install Miniconda by following the official Miniconda installation guide.

3. Create a Conda Environment
   Create and activate a Python 3.12 environment:

   conda create -n acp_designer python=3.12
   conda activate acp_designer

4. Install Dependencies
   Install the required packages:

   pip install -r requirements.txt

   or

   pip3 install -r requirements.txt

Web Application

No installation required! Access 'ACP-designer' full functionality via our web application:
👉 Web App Unfortunatelly only possible within Masaryk University intranet. Accessibility to the 'outside world' will be provided after security risks will be resolved

---

Running ACP_designer (Local Installation)

Generation of new ACPs

python run_cli.py --id 01 --nbatch 100

Generation of new ACPs + mutation of desired sequence

python run_cli.py --id 01 --nbatch 100 --mutate INWKGIAAMAKKLL

Command-Line Parameters

Customize runs with these parameters:

Parameter	Description	Options/Default
--id	run id to mark each run uniquely	integer value, always required
--nbatch	controls amount of generated peptides	100 (default)
--device	device	cpu (default), cuda, mps
--mutate	sequnce to mutate	string of amino acids, i.e: KKWLKA...

---

Success Story

We were able to design novel Mastoparan mutants with low hemolytic activity (RBC - red blood cell) and promising activity against human adenocarcinoma cells (A549):

---

Future Plans

Work in progress!