https://github.com/giotto-ai/molecule_bond_prediction

Predict scalar coupling in molecules
https://github.com/giotto-ai/molecule_bond_prediction

molecular-structures topological-data-analysis

Last synced: about 1 month ago
JSON representation

Predict scalar coupling in molecules

• Host: GitHub
• URL: https://github.com/giotto-ai/molecule_bond_prediction
• Owner: giotto-ai
• License: other
• Created: 2019-12-02T11:47:53.000Z (about 5 years ago)
• Default Branch: master
• Last Pushed: 2021-03-14T04:56:42.000Z (almost 4 years ago)
• Last Synced: 2023-03-06T12:57:01.490Z (almost 2 years ago)
• Topics: molecular-structures, topological-data-analysis
• Language: Python
• Homepage:
• Size: 5.68 MB
• Stars: 12
• Watchers: 3
• Forks: 5
• Open Issues: 2
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Predicting Molecular Bond Strengths with Topological Machine Learning

## What is it?

This repository showcases the core functionalities of

[giotto-learn](https://github.com/giotto-ai/giotto-learn), a Python library for

topological machine learning. The accompanying blog post can be found [here](https://medium.com/p/getting-started-with-giotto-learn-a-python-library-for-topological-machine-learning-451d88d2c4bc?source=email-22cbcf87e250--writer.postDistributed&sk=c5dffceb84a1108ea916f136f519258e).

This demo is based on the [Predicting Molecular

Properties](https://www.kaggle.com/c/champs-scalar-coupling/overview)

competition on Kaggle, where the task is to predict the bond strength between atoms in molecules.

## Getting started

The easiest way to get started is to create a conda environment as follows:

```

conda create python=3.7 --name molecule -y

conda activate molecule

pip install -r requirements.txt

```

## Results

The scoring function is described on Kaggle and is calculated as follows:







where:

* ![T](https://render.githubusercontent.com/render/math?math=T) is the number of coupling types

* ![n_t](https://render.githubusercontent.com/render/math?math=n_t) is the number of observations of type t

* ![y_i](https://render.githubusercontent.com/render/math?math=y_i) is the actual coupling value for this sample

* ![](https://render.githubusercontent.com/render/math?math=%5Chat%7By_i%7D) is the predicted coupling value for this sample

The figure below summarizes the results and gives a comparison of the results

with and without topological features.







## External code

The following Kaggle notebooks were used for this project:

* For non-topological features: https://www.kaggle.com/robertburbidge/distance-features 


* For plotting molecules (but adapted): https://www.kaggle.com/mykolazotko/3d-visualization-of-molecules-with-plotly

## Some related publications

To get an introduction to the application of topological data analysis to

machine learning, see:

* An introduction to Topological Data Analysis: fundamental and practical aspects for data scientists: https://arxiv.org/pdf/1710.04019.pdf

The idea to use topological data analysis for predictions on molecules is not new. Below you can find some interesting papers related to this:

* Persistent-Homology-based Machine Learning and its Applications – A Survey: https://arxiv.org/abs/1811.00252 (esp. section 5)

* Representability of algebraic topology for biomolecules in machine learning based scoring and virtual screening: https://arxiv.org/pdf/1708.08135.pdf

The following papers were used to get some inspiration for the feature creation:

* The Ring of Algebraic Functions on Persistence Bar Codes: https://arxiv.org/pdf/1304.0530.pdf

* A topological approach for protein classification: https://arxiv.org/pdf/1510.00953.pdf