A Deep Learning and Single-Cell Approach to Characterize Patient Response to Virus-Specific T-Cell Therapy
Open Access DepositedDownloadable Content
(i) understanding the relationship between the T-cell repertoire and gene expression and antiviral response, and (ii) predicting the likelihood of response to VST treatment based on clinical variables. The work outlined in this dissertation focuses on characterizing the clinical and cellular features which correlate with response to VST therapy in an effort to improve patient selection for future VST studies. In Chapter 2, we explore the single-cell RNA data in 4 patients who responded to hexaviral specific T-cell therapy, along with the 4 donor-derived products they received as part of a clinical trial (NCT03180216). In doing so we test and identify the best method for annotating T-cell samples, as most single-cell annotation methods focus on more general populations than T-cell subsets (i.e. identifying central and effector memory T-cells, cytotoxic CD8s instead of a general CD8 population, etc.). In Chapter 3, we apply these methods to assess patients from another clinical trial (NCT03475212), which used third party triviral products in the post-bone marrow transplant (BMT) setting. Beyond the RNA sequencing used in Chapter 2, here we also assess the expression of a small panel of proteins and relevant antigen specificity. In Chapter 4, we further explore the NCT03475212 patients, focusing on the pre-infusion clinical data, ranging from prior therapies given to original diagnosis to details of the bone marrow transplant. We develop a novel method for generating synthetic patient records and apply these records to train a deep neural network to predict the likelihood of response to VST therapy. This model is then tested on two other VST patient cohorts from collaborators at Cincinnati Children’s Hospital and Westmead Hospital. Taken together, this work furthers our understanding of what cellular and pre-clinical factors are relevant to a patient’s ability to overcome a viral infection upon treatment with VSTs. Virally infected immunocompromised patients are at severe risk of morbidity and mortality. The aims described in the coming chapters address the lack of in-silico models for the outcomes of VST therapy, which will predict how patients respond to VST treatment and how the immunome and transcriptome in T-cells affect such response. Using key clinical features as inputs to the model, the output provides clinicians with an indication of the patient’s likelihood of achieving viral clearance with VST therapy. Additionally, this work characterizes differences in the T-cell populations pre and post treatment in VST infusion patients to further understand potential variation in response to VST therapy, with the potential to incorporate these findings into a predictive model that could guide clinical decision making. In this work, we utilize a variety of T-cell analysis techniques (TCR-seq, scRNA-seq, flow cytometry), and apply the latest in machine learning and computational model development. In completing this work, we provide a critical predictive model to improve outcomes for immune compromised patients.
Immunocompromised patients, including transplant recipients, are at increased risk of severe infections and mortality from cytomegalovirus (CMV), Adenovirus (AdV), and Epstein-Barr virus (EBV). Viral infection is a frequent complication of hematopoietic stem cell transplant (HSCT), requiring antiviral therapy that is often complicated by toxicities and or resistance. Virus specific T-cell (VST) therapy has been an effective approach to treat viral infections in immunocompromised patients, including CMV, AdV, and EBV. Studies of VST therapy have shown 70-90% response rates, indicating the efficacy of the approach. However, variable antiviral responses remain a challenge, and currently there is no method to distinguish patients who will likely achieve viral clearance with VST compared to alternative therapies. Two critical gaps in the field are
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.