AHO

This is a repository for paper "Towards Data-driven Design of Asymmetric Hydrogenation of Olefins: Database and Hierarchical Learning". Here, you can find scripts used in this study.

👉️The full AHO dataset is available at AHO-Dataset.zip. 👈️

The online database server can be accessed at http://asymcatml.net/

Introduction

Asymmetric hydrogenation of olefins is one of the most powerful asymmetric transformations in molecular synthesis. Although several privileged catalyst scaffolds are available, the catalyst development for asymmetric hydrogenation is still a time- and resource-consuming process due to the lack of predictive catalyst design strategy. Targeting the data-driven design of asymmetric catalysis, we herein report the development of a standardized database that contains the detailed information of over 12000 literature asymmetric hydrogenations of olefins. This database provides a valuable platform for the machine learning applications in asymmetric catalysis. Based on this database, we developed a hierarchical learning approach to achieve predictive machine leaning model using only dozens of enantioselectivity data with the target olefin, which offers a useful solution for the few-shot learning problem in the early stage of catalysis screening.

Dependence

In order to run Jupyter Notebook for machine learning application demonstration, several third-party python packages are required.

python>=3.8.5
numpy>=1.19.2
pandas>=1.2.0
ase>=3.21.0
dscribe>=1.0.0
rdkit>=2019.09.3
openbabel>=3.1.0
scikit-learn>=0.23.2
mordred>=1.2.0
matplotlib>=3.3.2

We suggest using Anaconda to install the python 3.8.5 or higher version, as conda and pip together make the installation of these dependences much easier. All test are executed under Ubuntu 18.04, as the dscribe package currently only support Unix-based systems.

Installation of dependence

We suggest using Anaconda to prepare dependence as many packages are built-in Anaconda base environment. For those packages not built-in, you may input following commands to install them and follow the installation instructions.

conda install ase
pip install dscribe
conda install rdkit -c rdkit
conda install openbabel -c conda-forge
conda install -c rdkit -c mordred-descriptor mordred

Usage

Here we provide several tutorials in Jupyter Notebook format to demonstrate how to generate descriptors with provided reaction data, train machine learning model and use hierarchical learning approach to handle few-shot learning problem.

Dataset availability

You can find information about reaction of Asymmetric Hydrogenation over there.

How to cite

If the database or hierarchical learning is used in your publication, please cite as: Xu, L. -C.; Zhang, S. -Q.; Li, X.; Tang, M. -J.; Xie, P. -P.; Hong, X. Angew. Chem. Int. Ed. 2021, 60, 22804.

Contact with us

Email: hxchem@zju.edu.cn; licheng_xu@zju.edu.cn

---

📝 AHO Modification Change Log

Project Name: Asymmetric Hydrogenation of Olefins (AHO)
Goal: Fix Windows compatibility, resolve expired external dataset URLs, and improve code stability under modern Matplotlib and RDKit environments.
Test Environment: Windows 11/10 | Python 3.10 | RDKit | Scikit-Learn | Matplotlib

---

📂 Summary of Modified Files

1. Core Script & Utility Modifications

• examples/mlutils.py
  Fixed invalid LaTeX math syntax in drawregfig by replacing \itG with standard G (ΔΔG). Added try-except exception guard around plt.tight_layout().

---

2. Tutorial Notebook Fixes

• examples/01-generate_descriptors.ipynb
  Replaced Linux-specific bash commands (! wget, ! unzip) with cross-platform Python zipfile extraction logic.

• examples/02-train_simple_ML_model.ipynb
  Replaced %matplotlib notebook with %matplotlib inline. Added fallback logic to extract local datasets_for_fig_4.zip and generate dataset matrices (resolving expired asymcatml.net URLs).

• examples/03-hierarchical_learning.ipynb
  Wrapped from rdkit.Chem import Draw in try-except to prevent Windows GUI DLL load failures. Replaced ! cat /proc/cpuinfo with os.cpu_count(). Updated display.set_matplotlib_formats('svg') to standard set_matplotlib_formats('svg').

• examples/04-generate_correct_chiral_biaryl.ipynb
  Added os.makedirs('./biaryl', exist_ok=True) to ensure mol_1.sdf exports smoothly.

• examples/05-generate_correct_molecule_contain_ferrocene.ipynb
  Added try-except guard for Draw import and display. Included missing ferrocene template examples/Ferr/Ferr_std_reverse.sdf.

• examples/06-show_olefins_approaching_PCA.ipynb
  Replaced %matplotlib notebook with %matplotlib inline. Added Draw import guard.

---

3. Added Template & Preprocessed Datasets

• examples/Ferr/Ferr_std_reverse.sdf
  Generated reverse ferrocene 3D template file by flipping X coordinates of Ferr_std.sdf.

• examples/data/screening_desc_ensemble.npz
  Preprocessed descriptor ensemble dataset generated from local related_dataset_a.csv.

• examples/data/target.npz
  Preprocessed target dataset.

• examples/data/hierarchical_learning_set.npz
  Preprocessed hierarchical learning dataset.

---

🛠️ Key Improvement Areas

1. Cross-Platform Compatibility (Windows vs Linux)

• Problem: Original code used Linux-specific bash commands (! wget, ! unzip, ! mv, ! cat /proc/cpuinfo), causing syntax errors or missing command failures on Windows.
• Solution: Refactored logic using Python standard library modules (os, zipfile, sys) to ensure out-of-the-box execution across Windows, macOS, and Linux.

2. Expired External Data Source Fallback

• Problem: Original code relied on external laboratory server http://asymcatml.net/download/..., which is now unreachable (404 Error).
• Solution: Added automated extraction and RDKit descriptor fallback using local datasets_for_fig_4.zip, eliminating external server dependency.

3. Modern Matplotlib & RDKit Syntax Compatibility

• Problem:
  • Matplotlib math parser throws ValueError: Unknown symbol: \itG for legacy \itG LaTeX syntax.
  • %matplotlib notebook triggers IPython is not defined JavaScript error in modern JupyterLab.
  • RDKit Draw import fails on Windows environments missing GUI/Cairo C++ DLLs.
• Solution: Switched to %matplotlib inline, updated LaTeX \itG to G, and added try-except guards around Draw imports.