Taykhoom/siraj-mpra

Siraj MPRA measurements

A curated association-context view of the single-variant massively parallel reporter assay from Siraj et al., partitioned by assayed cell type. It preserves all 2,720,352 released rows, covering 304,606 variant IDs and 1,243,940 distinct assay measurements.

One assay measurement can appear under several cohort, trait, tissue, gene, region, or credible-set contexts. Those rows are scientifically useful and are not collapsed. pair_id, measurement_context_count, and is_measurement_representative let users recover one row per (variant, cell_type, library, ID) when needed.

The complete 119-column processed table remains reproducible with --full in the linked processing repository.

Cell-type names are physical access partitions, not machine-learning train/test assignments. Keep each pair_id together across association contexts, or group by variant for variant-held-out evaluation. The source release defines no canonical machine-learning split and no official evaluation metric.

log2_skew is the source log2Skew, the published paired-model coefficient. It is not the arithmetic difference alt_activity - ref_activity; the two activity coefficients were fitted separately.

Oligo sequence convention

sequence and alt_sequence contain the reference- and alternate-allele designed assay inserts. They do not contain the 15 bp synthesis adapters. For a full synthesized oligo, the official adapters were:

ACTGGCCGCTTGACG + stored_sequence + CACTGCGGCTCCTGC

If starting from a full synthesis sequence, trim sequence[15:-15]. Four official designs contain N padding; the dataset preserves those bases. The processing verifier confirms that allele 1 becomes allele 2 through the declared variant replacement. Sequence lengths span 73–200 bases and are not padded or truncated; indel alleles can change length.

Columns

The public table contains 31 selected and derived columns. Shared generic names are used only where their semantics match.

Columns	Meaning
source_index	Zero-based data-row index in core_mpra.txt.gz.
pair_id	Stable hash-derived identifier for (variant, cell_type, library, ID).
measurement_context_count	Number of released association-context rows carrying this same assay measurement.
is_measurement_representative	True on the first source row for each pair_id; filter this for one row per assay measurement.
sequence, alt_sequence	Official Supplementary Table 2 reference- and alternate-allele oligos.
label	Published expression-modulating-variant (emVar) call.
log2_skew, log2_skew_se, log2_skew_log_padj	Published paired-model alternate-versus-reference effect, standard error, and adjusted significance score. The last field contains 412 source-native positive infinities.
cell_line, assay_id, library	Assayed cell line, released result/window identifier, and oligo-library batch.
variant, hg38_variant, ref, alt	Source-intended hg19/GRCh37 variant coordinate, optional source-provided lifted hg38 metadata, and assayed alleles. The upstream release documents the hg19/GRCh37 intent, but cohort-wide coordinates were not independently revalidated locally. Source variant_hg38, published here as hg38_variant, was not locally re-derived.
ref_activity, ref_activity_se	Reference-allele reporter activity and standard error.
alt_activity, alt_activity_se	Alternate-allele reporter activity and standard error.
is_active	Published active-element call.
cohort, trait, region	Study/control source, phenotype, and locus region for this association row.
tissue, gene	GTEx tissue and gene context when available.
index_variant	Lead/index variant associated with the row.
credible_set_id	Released credible-set identifier.
pip	Posterior inclusion probability for this association context.
context_type	Released record class, such as credible-set, PIP-threshold, annotation, or control membership.

The complete --full output retains every processed source column, including the broader regulatory annotations and raw count summaries. It is not uploaded to this Hub repository.

The positive infinities in log2_skew_log_padj are preserved without clipping. Consumers that require finite inputs should mask that feature and retain an infinity indicator.

Processing and verification

The complete downloader, Parquet builder, source hashes, sequence checks, closed-world manifest verifier, and optional deterministic same-implementation reconstruction are available at:

https://github.com/TaykhoomDalal/Siraj-Processing/tree/main/MPRA

Stable repository-wide processing documentation: https://github.com/TaykhoomDalal/Siraj-Processing/blob/main/README.md

Citation

Siraj, L. et al. Functional dissection of complex trait variants at single-nucleotide resolution. Nature (2026). https://doi.org/10.1038/s41586-026-10121-6

License

This generated dataset combines CC BY-NC 4.0 Zenodo measurements with CC BY 4.0 Nature Supplementary Table 2 oligo sequences. Downstream use must satisfy both; license: other avoids presenting either component license as a universal relicensing. The processing implementation is MIT, and consulted upstream GPL-3.0 analysis code is not copied into the dataset. See the component matrix in the processing repository's NOTICE.md.