CARINA: Tissue-Resolved RNA-Binding Protein Grammar Models

A model framework trained on a proxy benchmark spanning 160 tissue-RBP combinations across eight human tissues. The dataset includes evaluation on 173 independent real-tissue CLIP datasets and was authored by Haoran Zhu, last updated in April 2026.

Use Cases

• Train tissue-transferable RBP binding predictors based on RNA sequence-structure graphs.
• Analyze tissue-resolved motif families and affinity landscapes based on the described context-calibrated grammar framework.
• Reconstruct locus-level binding architectures in extrapolation settings, such as the 15 TARDBP sites in the human brain NAT8L 3'UTR.
• Perform mechanism-oriented analyses at specific loci like MAPT based on interpretable model readouts.

Strengths

• Trained on a benchmark covering 160 tissue-RBP combinations.
• Evaluated on 173 independent real-tissue CLIP datasets.
• Model demonstrates transferability across eight human tissues.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

figshare

Collection Method

Model trained on heterogeneous CLIP resources and a GTEx-guided proxy benchmark.

Freshness

Last updated 2026-04-11 05:48:48; freshness should be verified.