The model takes as input single-cell ATACseq counts at a genic locus, aggregated over 500 bp tiles spanning 250 kb upstream/downstream and including the gene body, and uses regularized Poisson regression to predict the gene’s single-cell expression; both the scATACseq and scRNAseq readouts are obtained from multi-omic sequencing. The learned regression coefficients indicate the importance of each tile for predicting gene expression.
