Konstantin Krismer (GIFFORD LAB)

Identifying Neural Network Architectures for Genomics Prediction Tasks Using Simulated Data

Deep neural networks have achieved state-of-the-art performance on genomics prediction tasks, but they contributed little to the mechanistic understanding of the biology of regulatory elements, which is in part due to the complexity of the predictive models. To address this, we introduce seqgra, a deep learning pipeline that simulates data adhering to prespecified rules of a hypothesized model of genome regulation and identifies neural network architectures capable of recovering the rules behind the simulated data.

Connor Dobson (BIRNBAUM LAB)

Engineering Viruses to Decode and Reprogram Immunity

T cells use their antigen receptors to defend against pathogens and retain “memory” of the attack to bolster the immune defense against subsequent exposures. Understanding this recognition event remains a key unmet challenge in immunology. Current techniques used to determine their antigen specificity remain laborious and are limited to studying, at best, a handful of the millions of unique T cells found in a single person. Here, I discuss the development of a high throughput assay to determine T cell recognition at repertoire scale.