Multi-ancestry modeling improves fine-mapping resolution, protein prediction, and discovery for proteome-wide association studies

This Zenodo file collection contains TOPMed MESA-trained proteome prediction models built for PrediXcan, pQTL and fine-mapping output from TensorQTL SuSiE, SusieR, SuShiE, MultiSuSiE, and SuSiEx, and other supplemental information that belong to the research paper "Multi-ancestry modeling improves fine-mapping resolution, protein prediction, and discovery for proteome-wide association studies".

Proteome prediction models are titled by `trainingmethod_ancestry_snpinput.zip` and contain .db and .covariances.txt.gz files compatible with PrediXcan;

• Training Methods: Elastic Net (EN), Multivariate Adaptive Shrinkage (MASHR), Ultimate Deconvolution in R (UDR), Ensemble
• TOPMed MESA Ancestries: Combined Population (ALL, n=2953), European (EUR, n=1270), African (AFR, n=675), Hispanic/Latino (HIS, n=642), Chinese (CHN, n=366)
• SNP Inputs: Cis (cis), Cis fine-mapped (cisfm), Cis and Trans (cistrans), Cis fine-mapped and Trans fine-mapped (cistrans_fm)

`meta_cis_tensor_vs_susier_vs_sushie_vs_multi_vs_susiex.tsv` contains META cis-fine-mapped output from TOPMed MESA EUR, AFR, HIS, and CHN ancestral populations across 5 fine-mapping models - TensorQTL SuSiE, SusieR, SuShiE, MultiSuSiE, and SuSiEx.

TensorQTL SuSiE implementation ancestry-stratified cis- and trans-fine-mapped outputs;

• `cis_finemap_{ancestry}.tsv`
• `trans_finemap_{ancestry}.tsv`

`protein_coding_gene_boundaries.txt` contains left and right genomic coordinates used for pQTL mapping.

Code used to generate files can be found at https://github.com/ckrueger2/MESA_FM_PWAS_2026.