Machine Learning Methods for Biomolecular Modeling and Discovery 1200-UMPB-OG

1. Fundamentals of Machine Learning
1.1 Machine learning methods; neural network architectures; data encoding; training and validation.
1.2 Applications in biomolecular modeling and drug discovery.
1.3 Introduction to the representation of proteins, ligands, and chemical compounds. Overview of data formats such as SMILES, PDB, and FASTA.
1.4 Basics of cheminformatics; models describing chemical compounds, calculation of molecular properties (QSAR/QSPR).

Demonstration: Using the RDKit library in Python to parse SMILES and convert them into molecular graphs. Visualizing structures and calculating basic properties such as molecular weight, hydrogen bond count, and polarity.

2. Machine Learning in Biomolecular Modeling
2.1 Simple classification and regression models; predicting ligand binding to proteins.
2.2 Advanced network architectures: convolutional, recurrent, and graph neural networks.

Demonstration: Using the PyTorch library to predict small-molecule binding to cytochrome CYP1A2.

3. Generative Neural Networks
3.1 Theoretical introduction to generative neural networks and their applications in molecular modeling.
3.2 Generating new chemical molecules, optimizing drug properties, and analyzing chemical space.
3.3 Examples of generating chemical compounds and proteins using conditional models (cVAEs).

Demonstration: Using cVAEs to generate new molecules from a property focused space, such as toxicity or pharmacokinetics.

4. Protein Design Using ML
4.1 Concepts of protein design, including de novo design, sequence modifications, and practical applications.
4.2 Leveraging generative ML methods for generating new protein sequences and structures.
4.3 Machine learning methods used in protein design: RFDiffusion, AlphaFold, Protein MPNN.

Demonstration: Designing a globular protein using existing online models.