SenePy: Unveiling the Cell-Type Specific Landscape of Cellular Senescence through Single-Cell Analysis in Living Organisms (preprint)

Senescent cells accumulate in tissues with organismal age and contribute causally to multiple chronic diseases. In vivo senescent cell phenotypes are heterogeneous because cellular context and stressors vary by cell type and tissue. Due to the variability of senescence programs, there is no universal method to identify senescent cells and even widely used markers, such as CDKN2A, are not ubiquitous. Therefore, we interrogated the Tabula Muris Senis mouse single-cell aging atlas and an array of single-cell datasets from human donors that spanned many ages to find cell-specific signatures of cellular senescence. We derived 75 mouse and 65 human senescence signatures from individual cell populations. CDKN2A and other markers of senescence were overrepresented in these signatures but there were many novel senescence genes present at higher rates. Within individual cell populations, we observed multiple programs of senescence with distinct temporal and transcriptional characteristics. We packaged the signatures along with a single-cell scoring method into an open-source package: SenePy. SenePy signatures better recapitulate cellular senescence than available methods when tested on multiple in vivo RNA-seq datasets and a p16ink4a reporter single-cell dataset. We used SenePy to map the kinetics of senescent cell accumulation across 97 cell types from humans and mice. SenePy also generalizes to disease-associate senescence and we used it to identify an increased burden of senescent cells in COVID-19 and myocardial infarction. This work provides a significant advancement towards our ability to identify and characterize in vivo cellular senescence.

Language Models for the Prediction of SARS-CoV-2 Inhibitors (preprint)

The COVID-19 pandemic highlights the need for computational tools to automate and accelerate drug design for novel protein targets. We leverage deep learning language models to generate and score drug candidates based on predicted protein binding affinity. We pre-trained a deep learning language model (BERT) on ~9.6 billion molecules and achieved peak performance of 603 petaflops in mixed precision. Our work reduces pre-training time from days to hours, compared to previous efforts with this architecture, while also increasing the dataset size by nearly an order of magnitude. For scoring, we fine-tuned the language model using an assembled set of thousands of protein targets with binding affinity data and searched for inhibitors of specific protein targets, SARS-CoV-2 Mpro and PLpro. We utilized a genetic algorithm approach for finding optimal candidates using the generation and scoring capabilities of the language model. Our generalizable models accelerate the identification of inhibitors for emerging therapeutic targets.