Nanobody Diffusion: Amino Acid Sequence Generation using Diffusion Models

This project demonstrates how to use diffusion models for generating nanobody (single-domain antibody) amino acid sequences. It is designed as a research resource, providing a full pipeline from data preparation to model training, prediction, and visualization. Key features in this project include:

• Preparing and processing amino acid sequence data for machine learning
• Training a diffusion model to generate nanobody sequence generation
• Create new nanobody amino acid sequences from random data using model
• Scripts to visualize sequence data and model outputs

Results

_{AlphaFold predicted structure of generated nanobody sequence}

_{Amino acid frequency comparison between generated and test sequences}

_{Comprehensive frequency analysis across all amino acids}

How to install and run the Nanobody Diffusion project

The easiest way to get started is to follow these steps:

1. Clone this project

   git clone https://github.com/nathangendler/Nanobody_Diffusion
   cd Nanobody_Diffusion

2. Set up your Python environment

   • Install Python 3.8+ (recommend using Anaconda or Miniconda)
   • Create and activate a new environment:

     conda create -n nanobody_diffusion python=3.8
     conda activate nanobody_diffusion

3. Install dependencies

   • Dependencies can be found in the requirements.txt file

     pip install -r requirements.txt

4. Prepare your data

   • Place your amino acid sequence data (e.g., FASTA files) in the data/ directory and edit the path in diffusion/diffusion_train.py

5. Train the diffusion model

   • Run the training script:

     python diffusion/diffusion_train.py

6. Predict new sequences

   • Use the prediction script:

     python diffusion/diffusion_predict.py

7. Visualize results

   • Explore the output/ and diffusion/visual/ directories for visualization tools and results.

Future edits and modifications

This project is intended as a starting point for amino acid sequence generation research. You are encouraged to fork, clone, and modify the codebase for your own experiments. The modular structure makes it easy to:

• Adjust model architectures in diffusion/src/
• Extend training and prediction scripts
• Integrate new visualization tools