RNA3D-Decoy: A Large-Scale Benchmark Dataset of AI-Generated RNA Structure Models for Quality Assessment
Recent advances in AI- RNA structure prediction have increased the availability of candidate models, but their accuracy remains variable. Developing reliable quality assessment (QA) methods requires datasets that represent contemporary prediction errors across diverse targets and molecular contexts. Here, we present RNA3D-Decoy, a large-scale benchmark dataset comprising 103,549 decoys for 677…
RNA3D-Decoy presents a substantial collection of 103,549 decoys, created by five AI predictors, for 677 experimentally determined RNA targets. These targets encompass RNA-only structures, RNA structures with metal ions, and RNA-protein complexes. Each decoy is assessed using three metrics—TM-score, lDDT, and RMSD—against its experimental reference.
Additionally, RNA-protein interfaces undergo quality evaluation through DockQ scores. The dataset encompasses both RNA and protein sequence relationships to facilitate partitioning. Analyzes of RNA3D-Decoy expose variations in quality distributions across different levels and show target-dependent discrepancies among predictors.
Each predictor demonstrates the capability to generate the highest-TM-score model for some targets. Despite a broad pool of scores for RNA-protein complexes, the dataset also displays limited within-target score variation, underscoring the significance of target-level characterization. By amalgamating AI-generated decoys, standardized quality labels, and structured metadata, RNA3D-Decoy offers a valuable resource for enhancing global quality estimation, local error prediction, and candidate-model ranking.
This dataset, along with the supporting code, is accessible at https://github.com/HamiltonChenLab/RNA3D-Decoy.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
