Raw RNA-sequencing data for deriving BCR sequences from the NIH U01 project: Finding the optimal balance of immunotherapy efficacy and toxicity.

IrAEs convey substantial morbidity, incur considerable costs, and in some cases may preclude further use of immune checkpoint blockades. Recent studies indicate that (1) up to 80% of individuals on ICIs experience some form of irAEs; (2) ~35% of patients require corticosteroid treatments to mitigate these events; and (3) up to 20% terminate their therapy due to irAEs. As immunotherapy use extends from major centers to smaller, isolated, and less experienced community sites, predicting the development of irAEs and incorporation of this prediction into ICI treatment considerations becomes extremely valuable. Despite the substantial efforts invested in machine learning models for predicting ICI treatment responses, few studies have focused on predictive modeling for irAEs. We address this unmet need by leveraging the unique patient resources collected at UTSW and developing a BCR-based biomarker for irAEs. Moreover, irAEs can happen as short as a few weeks or as long as a few years after ICI treatment. We deployed our model longitudinally to track if the evolution of autoreactive BCRs align with the timings of irAEs.