Taykhoom/siraj-variant-pairs

Siraj variant-pair MPRA

Four-state measurements for nearby variant pairs from Siraj et al., joined to the official Nature Supplementary Table 18 RR, AR, RA, and AA oligo sequences. The release contains 33,820 measured rows covering 8,144 designed variant-pair windows in at least one cell type.

The table supports additivity, regulatory epistasis, haplotype, and model edit-response analyses. Public interaction_log2_skew is the source int_log2Skew; it belongs to the normalized state-effect layer, not the raw activity contrast.

The released normalized values satisfy interaction_log2_skew = altalt_log2_skew - altref_log2_skew - refalt_log2_skew.

Cell-type names are physical access partitions, not machine-learning train/test assignments. For strict variant-disjoint evaluation, group by component_id; for pair-held-out evaluation, group by variant_pair_group_id. pair_id is only the exact center/window design. The source release defines no canonical machine-learning split and no official evaluation metric.

Oligo sequence convention

sequence, altref_sequence, refalt_sequence, and altalt_sequence contain the designed assay inserts for the four states, with no synthesis adapters. A full synthesized oligo was:

ACTGGCCGCTTGACG + stored_sequence + CACTGCGGCTCCTGC

If starting from a full synthesis sequence, trim sequence[15:-15]. sequence_geometry_valid confirms the complete four-transition cycle, and sequence_orientation records direct or reverse-complement orientation. State lengths span 181–200 bases; indel alleles are preserved, and 994 measured designs have unequal lengths among their four states.

State names

State	Variant 1	Variant 2
refref / RR	reference	reference
altref / AR	alternate	reference
refalt / RA	reference	alternate
altalt / AA	alternate	alternate

Columns

The public table contains 45 selected columns. The complete 118-column source table remains reproducible with --full in the linked processing repository.

Columns	Meaning
source_index	Zero-based data-row index in haplos_master_table.txt.gz.
pair_id	Stable hash-derived identifier for the variant pair, center, and window.
ordered_variant_pair_id	Canonical ordered (variant1, variant2) identity, independent of center, window, cell line, and library.
variant_pair_group_id	Canonical unordered biological pair identity for pair-held-out evaluation.
component_id	Connected component of the shared-variant graph; use this for strict variant-disjoint train/test evaluation.
sequence, altref_sequence, refalt_sequence, altalt_sequence	Exact official Supplementary Table 18 RR, AR, RA, and AA oligos.
label	Published interaction emVar call.
interaction_log2_skew, interaction_log2_skew_se, interaction_log2_skew_padj	Published normalized double-edit interaction, standard error, and adjusted significance score.
cell_line, library	Assayed cell line and oligo-library batch.
variant1, variant2	Source-intended hg19/GRCh37 design coordinates for the two variants. The upstream release documents that assembly intent, but cohort-wide coordinates were not independently revalidated locally.
center_variant	Which variant was used to center the oligo.
window	left, middle, or right placement of the pair in the oligo.
distance	Source-released design distance. It is not independently recomputed from locally revalidated loci.
{state}_activity, {state}_activity_se	Absolute reporter activity and standard error for all four states.
altref_log2_skew, refalt_log2_skew, altalt_log2_skew	Each non-reference state relative to refref in the paired normalized model.
altref_log2_skew_se, refalt_log2_skew_se, altalt_log2_skew_se	Standard errors for those normalized state effects.
{state}_is_active	Published activity calls for all four states.
altref_label, refalt_label, altalt_label	Published expression-modulating-variant calls for the three non-reference states.
any_active, any_emvar	any_active is the OR of altref, refalt, and altalt activity and intentionally excludes the refref/RR baseline. any_emvar likewise summarizes non-reference states.
sequence_orientation	Orientation in which all four declared allele transitions validate; all released designs are direct.
sequence_geometry_valid	True only when all four required state transitions match the declared alleles.

The 2,522 unordered biological pair groups form 1,832 connected components over 4,080 variants. Exact repeated four-state sequence cycles never cross a variant_pair_group_id.

The complete --full output additionally retains raw count summaries, sequence lengths, regulatory annotations, and every other processed source column. It is not uploaded to this Hub repository. Its source field any_active_update includes all four states, unlike curated any_active.

Processing and verification

The complete downloader, Supplementary Table 18 join, four-state geometry checks, source hashes, closed-world manifest verifier, and optional deterministic same-implementation reconstruction are available at:

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

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 18 oligo sequences. Downstream use must satisfy both; license: other avoids flattening the mixed rights into one universal claim. 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.