December 1, 2024

Antigen Processing Prediction Acceleration

GPU-accelerated modeling of antigen processing to predict CD4+ T-cell epitopes.

Before a T cell can recognize an antigen, that antigen must be processed and presented as a
peptide on an MHC molecule. Modeling this antigen processing pathway lets us predict which
peptides become CD4+ T-cell epitopes, yet the underlying biophysical computations are
expensive. This project makes epitope prediction practical by accelerating the core
calculations on the GPU and packaging them into an easy-to-use suite.

Faster conformational stability

CD4+ epitope prediction depends on conformational stability calculations that dominate the
runtime. We re-engineered this computation for the GPU and achieved large speedups:

(BIBM 2024) International Conference on Bioinformatics and Biomedicine
Jiarui Li, Samuel J. Landry, Ramgopal R. Mettu

Accelerated sampling

The same pipeline relies on Markov Chain Monte Carlo sampling. We built a GPU
implementation that delivers substantial gains over the CPU baseline:

(AIMLSys 2024) Proceedings of the 4th International Conference on AI-ML Systems
Jiarui Li, Samuel J. Landry, Ramgopal R. Mettu

An integrated suite

Finally, we packaged antigen processing likelihood and epitope prediction into a single
application so others can run the full workflow end to end:

(Tech. Rep. 2026) Technical Report (Application)
Jiarui Li, Marco K Carbullido, Jai Bansal, Samuel J. Landry, Ramgopal R. Mettu

All related publications are listed below.

Related Publications

(BIBM 2024) International Conference on Bioinformatics and Biomedicine
Jiarui Li, Samuel J. Landry, Ramgopal R. Mettu
(AIMLSys 2024) Proceedings of the 4th International Conference on AI-ML Systems
Jiarui Li, Samuel J. Landry, Ramgopal R. Mettu
(Tech. Rep. 2026) Technical Report (Application)
Jiarui Li, Marco K Carbullido, Jai Bansal, Samuel J. Landry, Ramgopal R. Mettu

This post is written by Jiarui Li, licensed under CC BY-NC 4.0.

#Immunology #Antigen Processing #Algorithm Acceleration #Computational Biology