Cameron Tofani presented her work on biological pathway completion using transformers at the 2026 SMU Research and Innovation Week.

Abstract: Biological pathways describe how proteins work together inside a cell to carry out important processes such as metabolism and signaling. Many biological pathways are incomplete – reactions are known, but the enzymes (proteins) catalyzing them have not been identified. These missing enzymes create gaps known as pathway holes.

Filling pathway holes is critical for understanding disease mechanisms and identifying drug targets. This project investigates whether a BERT-based transformer model can predict candidate proteins for reactions where the enzyme is unknown.

Proteins and reactions are represented as tokens in sequences, allowing the model to learn biological context the way language models like BERT learn word context.

Experimentally identifying missing enzymes is slow and expensive. A computational approach could rapidly narrow the search space, which would accelerate disease research and drug discovery.

This work was supported by the Departmental Summer Seed Grant (DSSG), Department of Computer Science.

Poster

poster