Organ-specific prioritization and annotation of non-coding regulatory variants in the human genome

TLand_scores.zip: Scores for 2 million trait-associated variants from the GWAS catalog

TLand.zip: Data to reproduce model evaluations and GWAS catalog analyses

ukbb_data.zip: Data to reproduce UK BioBank analyses

caqtl_data.zip: Data to reproduce Wenz et al. 2025 caQTL analyses

models.zip: Final TLand, TLand light, and TLand lightest model objects saved as pickle files

Please see Fig1a to determine which model to use:

We defined organs with >=100 available ChIP-seq assays as “high data availability” and those with <100 assays as “low data availability”. We observed that the TLand (full) model consistently outperformed the TLand light model in organs with high data availability (Fig 1a and c; Supplementary Fig 5), while the light model surpassed the full model in organs with low data availability. For example, the organ-specific TLand light model was the best model when holding out the embryo organ (AUPR 0.639, AUROC 0.774). Those findings indicate that TLand light models are suitable for predicting regulatory variants for organs with low data availability while TLand (full) models are more suitable for organs with high data availability.