• Material for cheminformatics tutorial of CBI 2025.
• We will try to build GCN models and try to fine tune of foundation moldel.
• Please ignore warnings when you run the code on jupyter.
• 今回のチュートリアルにおいて、学習のEpochは非常に少なく設定してあります。GPUが利用できる環境下で参加される場合は適宜Epochsを増やしていただいて大丈夫です。CPUのみの環境で参加される方は、チュートリアル中は流れを把握していただく形にとどめていただくのが良いかと思います。

• 以下のコマンド例はBashでのテストをしました。Mac zsh環境で作業される場合Bashのコメントをうまく処理できない可能性があります。その場合以下のコマンドを実行してから作業して下さい。
• (stack over flow)[https://stackoverflow.com/questions/11670935/comments-in-command-line-zsh]

setopt interactivecomments

• Install git
• Gitのインストールをします。OSに応じて適宜コマンドを変更して下さい。
• Following example code is used for Linux

apt-get install git

• Get materials for the tutorial
• 今回のチュートリアルで使うコードをGithubからCloneします。Cloneができない場合はZipファイルとして入手し、利用していただいても大丈夫です。
• You can use git clone with SSH or Git CLI. Following code is an example.

git clone https://github.com/cbi-society/cheminfo_tutorial_20251027_pub.git
cd cheminfo_tutorial_20251027_pub 

• Install miniforge.
• After installing miniforge, mamba commad will be available. Please check more deitails in original documentation
• Miniforgeをインストールします。既にご自身の環境下に同様のものをインストールされている場合はこの作業はスキップしていただいて構いません。mambaはcondaより高速に動作するので、condaしか使えない場合はmambaを使えるようにすることをお勧めいたします。
• Following example code is used for Linux

curl -L -O "https://github.com/conda-forge/miniforge/releases/latest/download/Miniforge3-$(uname)-$(uname -m).sh"
bash Miniforge3-$(uname)-$(uname -m).sh

• Make conda enviromnet for the tutorial
• You can use your own favorite env name

conda create -n cbi2025 python=3.12

• Install required packages
• If you would like to use jupyter notebook, run mamba install -c conda-forge notebook
• zarr >= 3.0 cause error, so we should install zarr<3.0
• You can check installed packegs like $ conda list | grep zarr

conda activate cbi2025
mamba install -c conda-forge datamol molfeat medchem splito
mamba install -c conda-forge jupyterlab
mamba install -c conda-forge notebook
mamba install -c conda-forge auroris polaris openpyxl gcsfs umap-learn ipywidgets
mamba install -c conda-forge chemprop tensorboard bokeh
mamba install -c conda-forge zarr"<3.0"
pip install -U 'ray[tune]'
pip install hyperopt

• Install MolE in another environmet
  • We have some issues for installation MolE to MacOS because some packages such as triton aren't supported MacOS. So we are planining to work with chemprop mainly.
  • NOTE: Please comment out triton==3.0.0 in requirements/main_3.10.txt or requirements/main_3.9.txt if you are a mac user. Because triton supports only Linux.

# suitable python version is 3.9 or 3.10
cd ~/your_own_path
git clone https://github.com/recursionpharma/mole_public.git
cd mole_public
mamba create -n mole python=3.10
conda activate mole
mamba install -c conda-forge jupyter
# For Mac or CPU only:
pip install -r requirements/main_3.10.txt

# For CUDA:
pip install -r requirements/main_3.10_gpu.txt

pip install -e .

NOTE: If you are a mac user consider to use PYTORCH_ENABLE_MPS_FALLBACK=1 as environmental variable to avoid issues between torch and M1 processors. You can do it by typing (following example is bash as a terminal):

echo "export PYTORCH_ENABLE_MPS_FALLBACK=1" >> .bashrc

• Download pre-trained model(MolE_GaucaMol_27113.ckpt) checkpoint file from following URL.
• The size of this ckpt file is about 1GB. Please download the file in advance.ckpt file link

• To check the installation please run following command.

$ conda activate cbi2025
$ python check_deps.py test1.py

$ conda deactivate

$ conda activate mole
$ python check_deps.py test2.py

• Chemprop: A Machine Learning Package for Chemical Property Prediction
• chemprop_githubrepo
• Descriptor-based Foundation Models for Molecular Property Prediction
• MolE: a foundation model for molecular graphs using disentangled attention
• Recursion news
• MolE_githubrepo
• github proxy settings example
• wget proxy settings example
• conda proxy settings example

Please check the documentation above. RDKit>= 2025.03.5 will be required for tutorial by Greg. CBI2025 conda environment might have current version of RDKit

Please post issue if you find

ChemProp: MIT, MolE: Attribution-NonCommercial 4.0 International

Please post question to (discussion)[https://github.com/cbi-society/cheminfo_tutorial_20251027_pub/discussions].

• Following OS are testd
  • Ubuntu 24.0
  • WSL (Ubuntu22.0)
  • MacOS
  • Windows 11

• Taka@iwatobipen
• Kazufumi Ohkawa
• Koichiro Arai
• Natsumi Miyano
• Kazutoshi Takahashi