What is the next SARS-CoV-2? And how can we design a vaccine to protect against it? A key aspect of designing universal vaccines is to appropriately model the human immune system. In this project, we will develop statistical population models of the human immune system.
In particular, in this project, we will focus on data-driven modeling of the human leukocyte antigen (HLA), which is the part of the immune system which recognizes infected cells to be destroyed. A challenge with such a data-driven approach is that improved biological experiments have changed both the resolution of available data.
In this project, we will develop novel statistical and machine learning approaches that leverage this heterogeneous, multi-resolution data to build robust HLA population models.
Data
We will use publicly available HLA data from the Allele Frequency Net Database (AFND). This database includes both new, high-quality datasets with personalized HLA information as well as very large, traditional datasets from organizations like the US National Marrow Donor Program and the Shanghai Red Cross Marrow Donor Registry. In total, AFND includes immune system data on millions of individuals.
Methods
Due to the large dataset, as well as its semi-structured nature, we will investigate both Bayesian statistical models as well as generative neural network methods. In particular, we will apply and develop multi-resolution learning algorithms to account for the differences in data quality over time.
Evaluation
We adopt a holistic evaluation framework for this work. This will entail both quantitative strategies, such as accuracy and correlation with hold-out datasets, as well as more qualitative approaches, in which we will investigate the effects of models on downstream vaccine design tasks.
Outlook
Despite advances in medicine, many infectious diseases remain threats around the globe. An integral component of designing vaccines to address these threats is appropriately modeling the human immune system for different populations. The population models of the immune system built in this project will serve as the basis for numerous future vaccine designs.
Contacts
Main supervisor Brandon Malone brandon@oncoimmunity.com