BlueSTARR is a neural network trained on STARR-seq data that takes genomic sequence as input and predicts STARR-seq activity as output. This version of BlueSTARR was trained on a STARR-seq experiment done in K562 cells, and takes a 300 bp sequence as input. Variant predictions are made by running the model twice (once for the reference allele and once for the alternate allele) and computing the predicted log2 fold change in STARR-seq activity for the change from reference to alternate.