Datasets

The 20 molecular-function labels selected for the benchmark, assigned to proteins encoded by each transcript from Gene Ontology annotations.

The 20 biological-process labels selected for the benchmark, assigned to proteins encoded by each transcript from Gene Ontology annotations.

The 20 cellular-component labels selected for the benchmark, assigned to proteins encoded by each transcript from Gene Ontology annotations.

Isoform-level mean ribosome load measured in human cells with isoform-resolved ribosome profiling.

Mean ribosome load for randomized 5' UTRs attached to an eGFP reporter, including modified-RNA conditions.

Mean ribosome load for assayed 5' UTRs attached to an mCherry reporter.

Mean ribosome load for designed 5' UTR constructs from the Sample et al. reporter assay.

Mean ribosome load across reporter constructs whose 5' UTR lengths vary.

Ribosome load and RNA half-life measured together on a small collection of synthetic mRNA constructs.

Mean translation efficiency from a paired Ribo-seq and RNA-seq atlas, retaining the transcript with the highest translation efficiency for each human gene.

Mean translation efficiency from a paired Ribo-seq and RNA-seq atlas, retaining the transcript with the highest translation efficiency for each mouse gene.

A binary label indicating whether each assayed or curated candidate showed internal ribosome entry site activity.

Proximal isoform usage across synthetic 3' UTR alternative-polyadenylation windows.

Human transcript half-life aggregated from time-course RNA sequencing after transcription inhibition.

Mouse transcript half-life aggregated from time-course RNA sequencing after transcription inhibition.

Reporter stability for native, naturally varied, and designed 3' UTR fragments in the BTV EGFP transcript context in Jurkat cells.

Evaluation: L2 embedding VEP is unsigned. Fit ridge or OLS to alternate-minus-reference embeddings for signed effects; this route trains on labeled examples.

Reporter stability for native, naturally varied, and designed 3' UTR fragments in the BTV EGFP transcript context in BEAS-2B cells.

Evaluation: L2 embedding VEP is unsigned. Fit ridge or OLS to alternate-minus-reference embeddings for signed effects; this route trains on labeled examples.

Twenty separate binary classification targets; a positive target means eCLIP detected binding by that RNA-binding protein on the transcript in K562 cells.

Twenty separate binary classification targets; a positive target means eCLIP detected binding by that RNA-binding protein on the transcript in HepG2 cells.

Twenty separate binary classification targets record which frequent human miRNAs were observed among transcripts with at least one selected interaction.

Access: Raw rebuild only. The data are not redistributed; installation of the dev extra and GenomeKit is required to process the original MirTarCLASH download.

One label per annotated Human Protein Atlas compartment for the protein encoded by each transcript.

One label per APEX-seq compartment; a label is positive when its measured proportion is at least 0.125.

Access: Raw rebuild only under the Elsevier source license. Installation of the dev extra and GenomeKit is required.

Three labels marking normalized coverage below the 33rd percentile in chromatin, cytoplasm, and polysomes in K562 cells.

Pathogenicity prediction for 5' and 3' UTR variants associated with Mendelian disease.

Pathogenicity prediction for 5' and 3' UTR variants associated with complex traits.

Manually reviewed pathogenic and benign variants in human 5' UTRs.

Manually reviewed pathogenic and benign variants in human 3' UTRs.