Supervision: Daniel Probst
Semester project (master)
Summary. The wavefunction collapse algorithm is a conceptually simple generative algorithm inspired by wavefunction superposition . The algorithm has been used for procedural level and model design [2,3] and explored in the context of discriminative learning . Furthermore, it has been extended to graph generation . De novo molecular design is developing as a sub-field of chem- and bioinformatics concerned with sampling molecules from a conceptual chemical space . Recently, generative machine learning has led to an explosion in available methods capable of exploring this chemical space while imposing constraints such as drug-likeness or synthetic accessibility [7,8]. However, these models rely on large amounts of training data, potentially introducing unknown biases and largely limiting the scope of generated samples to already explored areas of the chemical space. A wavefunction collapse algorithm-based approach has the potential to solve this problem while being both interpretable and computationally efficient.
Goals. As a molecule is generally represented by its molecular graph, which may also include the spatial coordinates of the atoms, de novo molecular design can be posed as a graph generation problem. The main goal of this project is to explore the wavefunction collapse algorithm in the context of de novo molecular design, focusing on its extension to graphs and discriminative learning. Specifically the project aims to achieve the following:
- Implementing the Python scripts necessary to facilitate the application of the wavefunction collapse algorithm to molecular graphs.
- Training and testing a discriminative learning-based generator capable of creating molecular graphs with drug-like properties.
- (Bonus) Extending the approach to the tertiary structure of biological macromolecules, such as proteins.
Finally, I expect the work to be published in a scientific journal or presented at a conference.
Profile. You are a computer scientist or mathematician with an interest in the natural sciences or a chemist or biologist with an interest in computer science. Experience in programming (Python) is required.