Source data for Leshem et al., Long-read RNA sequencing in primary lung cell types reveals principles of nonsense-mediated decay.The starting data for every analysis in the paper — the processed quantification, annotation and model outputs that the analysis code reads and cannot produce for itself. Paired short-read and PacBio Iso-Seq sequencing of four primary human lung cell types (AT2, LAE, FB, MV) under SMG1 inhibition versus vehicle: 26 samples, 13 donor-matched pairs.Download the twelve data files, 695 MB — that is every input the code needs. The two files of about 8 GB each are the packaged software environment the analyses ran in, and are optional. Twenty further files, 27.3 GB, hold the model's full DeepSHAP attribution arrays, from which the record's DeepSHAP tables were computed; they are optional too.README.md, included in this record, has the full instructions: how to arrange the files, what each one is, and what may trip you up. Analysis code: github.com/peter4244/nmd_lung_longread_reproduction (DOI 10.5281/zenodo.21897099). Raw sequencing reads are in NCBI GEO under GSE329233.