Machine Learning in Drug Design (2024)

This repository contains course materials for the course "Machine Learning in Drug Design." In the labs directory, there are materials covering the following topics:

1. Python and machine learning basics (revision)
   • Textual representations of molecules: SMILES
   • Vector representations of molecules: descriptors and fingerprints
   • Introduction to RDKit
   • Classical ML models: Linear Regression, Random Forest, Support Vector Machines
2. Exploration of publicly available small molecule datasets
   • ChEMBL database of bioactive molecules
   • ZINC database of purchasable molecules
   • PubChem database of chemical information about small molecules
   • Exploratory Data Analysis (EDA)
   • Quantitative Structure-Activity Relationship (QSAR) and Virtual Screening (VS)
3. Graph neural networks
   • Neural networks architectures and training
   • Molecular graphs, atomic featurization
   • Message passing neural networks
   • Graph convolutional neural networks
   • Explainability: Grad-CAM
4. Molecular docking
   • Molecular data formats: SMI, SDF, MOL2, PDB
   • Force fields
   • Protein folding
   • Molecular docking with AutoDock Vina, Smina, QuickVina
   • Interaction fingerprints
   • Pharmacophores
5. Deep generative models
   • Autoencoders
   • Recurrent neural networks
   • SMILES generators: ReLeaSE and REINVENT
   • Graph-based generative models: JT-VAE
   • Reinforcement learning and Bayesian optimization for molecular property optimization
6. Protein deep learning
   • Simplified protein graph representations
   • Voxel grid representation
   • Mesh representation for encoding protein surface
   • 3D convolutional neural networks for encoding proteins
7. Uncertainty prediction
   • Aleatoric and epistemic uncertainty
   • Conformal prediction

Lectures

The lecture slides and notes are in the lectures directory. The machine learning lecture is deployed online at:

gmum.github.io/mldd24/lectures/ml-basics-for-drug-design.html

About us

GMUM (Machine Learning Research Group) is a group at the Jagiellonian University working on various aspects of machine learning, and in particular deep learning - in both fundamental and applied settings. The group is led by prof. Jacek Tabor.

Some of the research directions our group pursues include:

• generative models: efficient training and sampling; inpainting; super-resolution,
• theoretical understanding of deep learning and optimization,
• natural language processing,
• drug design and cheminformatics,
• unsupervised learning and clustering,
• computer vision and medical image analysis.

We are hosting machine learning seminars that are open to the public. You can check the schedule on our website and join online (links posted on our Facebook). You can also add seminar info to your Google calendar.

Environment Setup

Python will be used throughout the course. The environment setup steps are shown below:

1. Install miniconda following the instructions for your operating system.
2. Download this repository: git clone https://github.com/gmum/mldd24.git.
   • You need to have Git installed.
3. Install environment from the YAML file: conda env create -f environment.yml (or conda env create -f environment-gpu.yml for the GPU version).

Important! If you would like to use your GPU to train neural networks, update the line pytorch-cuda={cuda version} in the environment-gpu.yml file. The current CUDA version is 12.1, but you should check your graphics card compatibility first.

Grading scheme

The final course grade is equal to the project grade, if the laboratories have been passed.

The laboratory grade is based on the assignments turned in. Class attendance is mandatory and two unexcused absences are allowed. More unexcused absences will result in failing the course (NZAL).

percentage	grade
>90%	5
>80%	4.5
>70%	4
>60%	3.5
>50%	3
≤50%	2

Automatic grader available at mldd.matinf.uj.edu.pl