AI, the Immune System & Polygenic risk

I am excited to host the 8th Meetup of AI in Genomics!

Note that this event will be ONLINE!

To register, you MUST use the following link (seats are limited!):
https://www.getwebinar.co.il/lp/immunai20200930/?your-ref=ai_in_genomics_meetup

Schedule:

18:00 - Rebecca Herbst (Computational Biologist @Immunai)

"Studying dysfunction of tumor specific CD8+ T cells longitudinally across tumor development using single cell analysis"

During chronic antigen stimulation, CD8+ T cells lose their effector potential. However, because much of our understanding of anti-tumor T cell dysfunction comes from shorter term transplantable models, it is challenging to determine how this dysfunction evolves over time in tumors. Using single-cell RNA-Seq and functional assays in an autochthonous model of lung adenocarcinoma, we demonstrate that anti-tumor CD8+ T cell dysfunction is dynamic and heterogeneous. Applying probabilistic models to a continuous space of cellular states, we identified diversity within tumor specific TCF-1+ CD8+ T cells, a population known to retain superior functionality, which was dynamic over tumor development. Furthermore, blocking egress from the draining lymph node shifted the composition of TCF-1+ CD8+ T cells in tumor-bearing lungs, demonstrating that the lymph node is a reservoir of more functional anti-tumor CD8+ T cells. Collectively, our results provide insights into TCF-1+ T cell biology with therapeutic implications for anti-tumor immunity.

19:00 - Dr. Omer Weissbrod (Harvard University)

"Predicting genetic disease risk across diverse human populations"

Polygenic risk measures our genetic predisposition to diseases like diabetes or schizophrenia. Predicting polygenic risk is a key goal of genetics research, as it allows identifying individuals at risk years in advance. However, polygenic risk predictions based on European training data suffer reduced accuracy in non-European populations, exacerbating health disparities.

I will present PolyPred, a Bayesian framework to predict polygenic risk in arbitrary populations by combining three ideas: (1) identifying the most clinically important features among millions of variants; (2) prioritizing important features using hundreds of biological annotations from external datasets; and (3) pooling data across multiple populations to find shared genetic patterns. I will show that PolyPred dramatically improves cross-population polygenic risk predictions over the state of the art, particularly in African individuals.

Bio: Omer is a postdoctoral fellow at Alkes Price's group at Harvard University. His research lies at the intersection of machine learning, statistics & genetics. Previously he was a postdoctoral fellow at Eran Segal's group at the Weizmann Institute of Science. He completed his PhD at the CS department of the Technion under the supervision of Dan Geiger, and at the statistics department of TAU under the supervision of Saharon Rosset. He was also a part-time researcher at the Machine Learning for Healthcare & Life Sciences group at IBM Research Haifa, where he worked on a variety of projects involving analysis of genetic & clinical data.