saarantras1/mpac-predictions

MPAC variant effect predictions

MPAC (Malinois with Parallel Aggregated Cross-validation) predicts cis-regulatory activity of 200 bp human sequences in K562, HepG2 and SK-N-SH, and the allelic skew caused by non-coding variants. This dataset holds the published predictions.

Identifying non-coding variant effects at scale via machine learning models of cis-regulatory reporter assays.

The paper is the source of truth for how these tables were made and what they mean. This is a mirror of Zenodo record 20694870 (version of concept DOI 10.5281/zenodo.15178434), uploaded so the tables are queryable without downloading the archives. File contents are unmodified.

The models that produced these predictions are at saarantras1/MPAC; the MPRA training data is at saarantras1/mpac-mpra.

Supplemental data and resources for "Identifying non-coding variant effects at scale via machine learning models of cis-regulatory reporter assays"

Including:

MPAC and comparison model predictions for:

Siraj 2026 (UKBB/BBJ/GTEx) MPRA (Supplementary Dataset 1)

DNase allele specific effects (Supplementary Dataset 2)

ClinVar (20260104) (Supplementary Dataset 3)

COSMIC (v98) (Supplementary Dataset 4)

GWAS tag SNVs + LD variants (Supplementary Dataset 6)

Haplotype analyses (Supplementary Dataset 7)

gnomAD (v3.1.2 SNPs and and small indels) (Supplementary Dataset 8)

Saturation mutagenesis of human promoters (Supplementary Dataset 9)

Saturation mutagenesis ENCODE dELSs (Supplementary Dataset 12)

Supplemental Tables for:

Recurrent COSMIC v98 emVar candidate drivers (Supplementary Dataset 5)

Promoter level function x constraint relationships (Supplementary Dataset 10)

Functional enrichments for promoter level function x constraint relationships (Supplementary Dataset 11)

GTEx phenocopying results (Supplementary Dataset 13)

All promoter and dELS seqlets (Supplementary Dataset 14)

Model hyperparameters (Supplementary Dataset 15)

All MPAC model artifacts (Supplementary Dataset 16)

Detailed descriptions for all predictions and supplemental tables can be found in 'File Descriptions.xlsx'

Names that differ from Zenodo

The gnomAD SNP files and the per-chromosome promoter saturation mutagenesis files are named *.vcf.gz in the deposit but are gzipped 14-column TSVs carrying no VCF header. They are renamed *.tsv.gz here, byte-identical, so the Hub's loader reads them. The gnomAD Indels/ files are genuine bgzipped VCFs with predictions in INFO and keep their names; they are not exposed as a queryable config.

Citation

@article{butts2025mpac,
  title   = {Identifying non-coding variant effects at scale via machine learning
             models of cis-regulatory reporter assays},
  author  = {Butts, John C. and Rong, Stephen and Gosai, Sager J. and
             Castro, Rodrigo I. and Noon, Mackenzie and Adeniran, Kehinde and
             Ghosh, Rohit and Sabeti, Pardis C. and Tewhey, Ryan and Reilly, Steven K.},
  journal = {bioRxiv},
  year    = {2025},
  doi     = {10.1101/2025.04.16.648420}
}

License

CC-BY-4.0, matching the Zenodo deposit.