ALLM-Ab: Active Learning-Driven Antibody Optimization Using Fine-tuned Protein Language Models

This repository contains the code for ALLM-Ab, a multi-objective antibody optimization framework using protein language models.

1. allmab_offline: Evaluation of active learning in offline settings using BindingGYM dataset
2. allmab_online: Implementation of active learning in online settings using Flex ddG

Setup

Installation

The easiest way is to install the dependencies listed in pyproject.toml using uv.

git clone https://github.com/your-username/ALLM-Ab.git
cd ALLM-Ab
uv sync

Optional: Flex ddG installation is required for the allmab_online component.

Data

The datasets can be obtained from BindingGYM:

git clone https://github.com/luwei0917/BindingGYM

Usage

allmab_offline

Example:

python al_run.py exps/outputs_ablang/0/greedy_0.0/dms_0_N-50_ini-1/config.yaml

allmab_online

Example:

cd allmab_online
python al_run.py configs/5A12_dual/ablang2/greedy_dual.yaml

Filling Mode[TODO]

A filling mode that supports general online active learning processes is available as follows.

cd allmab_online
python al_run_filling.py sample.yaml

Project Structure

• allmab_offline: Offline environment for protein binding simulation
• allmab_online: Online analysis using Flex ddG
• notebooks: Jupyter notebooks for analysis
  • reproduction_bindinggym.ipynb: Notebook for reproducing bindinggym results
  • reproduction_flexddg.ipynb: Notebook for reproducing flexddg results
  • analysis: Analysis notebooks
• results: Results from experiments
  • allmab_offline: Results from allmab_offline
  • allmab_online: Results from allmab_online

Citation

Furui K, Ohue M. ALLM-Ab: Active Learning-Driven Antibody Optimization Using Fine-Tuned Protein Language Models. Journal of Chemical Information and Modeling, Article ASAP, 2025. https://doi.org/10.1021/acs.jcim.5c01577