- This event has passed.
Ph.D. Defense: Elham Sherafat
November 23, 2022 @ 11:00 am - 12:00 pm EST
Title: Bioinformatics and AI methods for neoepitope prediction in personalized cancer immunotherapy
PhD Candidate: Elham Sherafat
Major Advisor: Dr. Ion Mandoiu
Associate Advisors: Dr. Mukul Bansal, Dr. Sheida Nabavi
Day/Time: Wednesday, November 23, 2022 , 11:00 AM
Location: Remote via Webex
Join link: https://uconn-cmr.webex.com/uconn-cmr/j.php?MTID=m990d2d5a8de74fe2b1bb2f38b20dda05
Webinar number: 2624 135 1351
Webinar password: eZDJSGqh728 (39357474 from phones)
Join by phone +1-415-655-0002 US Toll
Access code: 262 413 51351
Abstract:
Personalized cancer vaccines are emerging as one of the most promising approaches to immunotherapy of advanced cancers. These vaccines seek to boost the immune response against tumor neoepitopes generated by the somatic mutations carried by cancer cells. Since only a small proportion of these neoepitopes lead to tumor rejection, developing accurate methods for predicting tumor-rejection mediating neoepitopes is critical.
This work introduces bioinformatics and machine learning methods that efficiently analyze cancer patients’ sequencing data and addresses several challenges in generating effective personalized cancer vaccines. First, we describe GeNeo, a comprehensive web-based bioinformatics toolbox for genomics-guided neoepitope prediction, including tools for somatic variant calling from multi-technology exome sequencing data, variant validation by targeted resequencing, and neoepitope prediction. Second, we propose a semi-supervised method, positive-unlabeled Learning using AuTOml (PLATO), to further improve the sensitivity of somatic variant calling from exome sequencing data and peptide identification from MS/MS data. Finally, we introduce NeoRa, a machine learning approach to rank tumor-rejection mediating neoepitopes. Besides previously used HLA-I binding, presentation, and recognition features, NeoRa incorporates a novel differential immunogenicity index. We test NeoRa on data from immunization experiments with long synthetic peptides in mouse models. The results show promising predictive power in scenarios similar to those required for clinical use.
