morrislab/rna-stability-siegel

Overview

This dataset contains 3' UTR fragment measurements from the fast-UTR massively parallel reporter assay reported by Siegel et al. The source library contains 41,255 sequences tested in Jurkat T cells and BEAS-2B airway epithelial cells. Each configuration retains rows with a T4 stability target, a T4 effect target, or a reference needed to pair a measured effect. The library contains native human 3' UTR fragments, natural variants, and designed mutations of regulatory elements.

The source study inserted each test fragment downstream of EGFP in the BTV reporter plasmid. The sequence column reconstructs that assay transcript as:

BTV 5' UTR + EGFP ORF + BTV 3' UTR prefix
+ variable 3' UTR test fragment + BTV 3' UTR suffix

The resulting sequences are 1,448-1,468 nt. The target is reporter stability in this fixed construct, not the half-life of a complete endogenous transcript. The random 8-nt indexes used to distinguish experimental clones are omitted: each source target is aggregated across multiple independently barcoded clones, so no single barcode belongs to a sequence-level row.

This reporter-context revision preserves the original row count, row order, identifiers, targets, effects, parent links, descriptions, genes, and chromosomes. Only sequence, cds, and splice differ from the original fragment-only release.

This dataset is redistributed as part of mRNABench: https://github.com/morrislab/mRNABench

Data Format

Each cell-line parquet supports three views:

• Full-library regression uses target for every measured reference, natural-variant, or designed-mutant fragment.
• Reference-only regression uses target_reference, which is populated only when description is wild-type.
• Paired VEP regression uses target_effect, the measured alternate-minus-reference effect. parent_oligo_id identifies the corresponding reference row.

For supervised embedding VEP, mRNABench regresses target_effect on alternate_embedding - reference_embedding. Zero-shot likelihood VEP uses an alternate-minus-reference score. A custom signed embedding scoring function can also be supplied; an L2 norm measures magnitude but discards effect direction.

Configuration	Rows	T4 targets	Reference targets	Paired effects
jurkat	30,848	30,253	9,620	18,454
beas2b	7,613	7,613	2,533	2,231

Description of data columns:

• oligo_id: Unique source oligo identifier.
• gene: Gene symbol associated with the source 3' UTR.
• chromosome: Source chromosome on GRCh37/hg19.
• sequence: Reconstructed BTV EGFP reporter transcript containing the assayed 150-170 nt variable 3' UTR fragment.
• cds: Length-matched binary codon-position track for the 720-nt EGFP ORF.
• splice: Length-matched all-zero track because the reporter transcript is unspliced.
• target: T4 reporter stability, E4 / (E4 + E0), where Et = RNA_t / (RNA_t + DNA).
• target_reference: target for wild-type rows and null otherwise.
• target_effect: Raw T4 alternate-minus-reference stability effect.
• parent_oligo_id: Reference oligo identifier for paired analysis.
• description: wild-type, natural-variant, or designed-mutant.

The source tables do not retain RefSeq transcript accessions. Many fragments are compatible with multiple isoforms, so no transcript identifier is inferred. Random row splitting is not appropriate because neighboring source windows overlap and reference/alternate sequences can differ by only one base. mRNABench uses chromosome-grouped splitting.

Reporter Context Provenance

The BTV transcript reconstruction matches the construct published with the Saluki analysis of the Siegel fast-UTR data:

• Repository commit: https://github.com/vagarwal87/saluki_paper/tree/3aa4e56a19bbbf87ac9f5f0a251098cf749ad6bc/Fig6_S7/Siegel_testSet
• Construct file: BTV_construct.txt.gz
• Decompressed construct SHA-256: a237c2b44d94d1e3b007aaea643eea06bbe320368639ad4983541aea600ae1dc
• BTV reporter plasmid: Addgene #84771 https://www.addgene.org/84771/

The published reconstruction contains a 161-nt BTV 5' UTR, the 720-nt EGFP ORF, a 45-nt 3' UTR prefix, the variable test fragment, and a 372-nt 3' UTR suffix. The EGFP ORF and constant transcript sequence also match the public BTV vector map. The original fast-UTR cloning protocol used MluI and SbfI sites in the BTV multiple cloning site.

Original fast-UTR method: https://doi.org/10.1038/nbt.2851

Measurement Source

The sequence-level measurement files are pinned to commit 4aaeb7e97c5ec093c356f0564f96d87887ee9ab7:

• sequence_level_data_Jurkat.csv
  SHA-256: 4b3e1b3452a7753d8308e3eb1c8847c3f2692ff2a08d699ea1e072d52f150f66
• sequence_level_data_Beas2B.csv
  SHA-256: 39b8cd5fad482e3ddec5b07f6975ff8447b50976c134de966c6e285a520b0ca8

The paper and supplementary materials are licensed under CC BY 4.0. The source GitHub repository does not specify a separate data license, so this repository is marked license: unknown.

Please attribute:

Original paper: https://doi.org/10.1093/g3journal/jkab404
Original dataset source: https://github.com/david-a-siegel/AU-Rich-Elements
Citation: Siegel DA, Le Tonqueze O, Biton A, Zaitlen N, Erle DJ. Massively parallel analysis of human 3' UTRs reveals that AU-rich element length and registration predict mRNA destabilization. G3. 2022;12(1):jkab404. https://doi.org/10.1093/g3journal/jkab404