Joel Joseph
Computer science
Hometown: Gilbert, Arizona, United States
Graduation date: Fall 2025
FURI | Spring 2025
Optimizing TCR Embeddings: Reducing Data Requirements for Efficient Model Performance
catELMo, a deep learning-based embedding model for T-cell receptor (TCR) sequences, has significantly improved immunological tasks like TCR-epitope binding prediction. However, its reliance on large-scale datasets raises concerns about its efficiency and accessibility, especially in fields where data collection is resource-intensive. This project aims to determine the minimal dataset required to maintain strong predictive performance by systematically reducing training data and evaluating the impact on model accuracy. Additionally, we will explore a data selection strategy that prioritizes sequence diversity to retain performance with a smaller dataset. By training and testing catELMo on progressively smaller, strategically selected datasets, we will identify a threshold where further reductions degrade accuracy. Our findings will help optimize the TCR modeling, making deep learning-based immunological research more scalable and applicable to real-world settings where data is limited.
Mentor: Heewook Lee
It’s hip to be square.
Students presenting projects at the Fulton Forge Student Research Expo are encouraged to download this personal QR code and include it within your poster. This allows expo attendees to explore more about your project and about you in the future.
Right click the image to save it to your computer.