Current work
What is in progress right now, ahead of write-up.
Engineering thesis: conditional molecular design
Steering a de novo molecule generator with biomedical knowledge-graph embeddings (PrimeKG, ChEMBL) as the conditioning vector, and testing whether it beats simple target encodings on unseen targets.
knowledge graphsgraph neural networksgenerative modellingdrug discovery
Deep learning from scratch
Building a deep-learning-from-scratch foundation for ML research in drug discovery — implementing neural network architectures (MLP, CNN, Transformer) from first principles in NumPy, with hand-derived backpropagation, applied to molecular property prediction on QM9.
deep learningtransformersmolecular property predictionNumPy