NMD_orf_model_v5_4ct: deep-learning model of nonsense-mediated decay from ORF sequence context

Deep-learning model predicting nonsense-mediated decay (NMD) susceptibility from sequence around the start and stop codons together with per-ORF structural features, for Leshem et al., Long Read RNA Sequencing in Primary Lung Cell Types Reveals Principles of Nonsense-Mediated Decay. Covers training, evaluation, and attention- and SHAP-based interpretability. Code only. The trained checkpoint, its per-isoform predictions and the full interpretability export are in the Zenodo source-data record (DOI 10.5281/zenodo.21544336); the starting sequencing data are in NCBI GEO under GSE329233. The analysis code for the rest of the paper is nmd_lung_longread_reproduction.