Help Topics

• About RNArank
• Workflow
• lDDT definition modification
• Result in test sets
• Understanding the RNArank Output
• How to cite RNArank?
• Need more help?

About RNArank

   Workflow

RNArank is a novel deep learning-based approach to both local (nucleotide-specific) and global quality assessment of predicted RNA 3D structure models. Shown in Figure 1a, the RNArank works as follows. (1) extracting multi-modal features from a given RNA structure, (2) predicting intermediate contact and deviation maps using a Y-shaped residual neural network, and (3) calculating a final predicted quality score (pLDDT). RNArank can rank a set of RNA decoys to identify high-quality models, showing superior performance in benchmarks like the CASP15 and CASP16 experiments.

Figure 1. Overall architecture of RNArank. (a) The quality assessment pipeline of RNArank. (b) The multi-modal encoder. (c) The 3D CNN block. (d) The ResNet block used in RNArank.

   lDDT definition modification

To better reflect the structural properties of RNA, RNArank adapted the standard lDDT (denoted as lDDT_raw) formulation by using larger distance thresholds. This RNA-adapted lDDT (denoted as lDDT_RNA) shows a significantly improved correlation with three other well-established metrics (RMSD, TM-score_RNA, and GDT_TS).

Figure 2. The formulation of lDDT_RNA score. Where L is the total number of residues; d_ij and D_ij are the C4’ distance between i-th and j-th residues in the predicted and experimental structures, respectively; I() is the indicator function.

   Result in test sets

RNArank has been validated in test sets, including T24, CASP15 and CASP16 (Figure 2), which suggests that the automated assessment by RNArank are competitive to other methods.

Figure 3. Performance of RNArank and comparison with other methods. Comparisons of the top 1 lDDT_RNA (left) and RMSD (right) on T24, CASP15, and CASP16 datasets. Error bars represent 20% of the standard deviation.

Job submission

Job submission guide:
(1) Input an RNA structure into this test box
(2) Alternatively, you can upload the file storing structure.
(3) (optional) Provide your email address here.
(4) (optional) Assign your target name here.
(5) Choose whether to keep your results private.
(6) Click the "Submit" button to submit your job.

Notice: Due to computing resource limitation, we now allow no more than 20 running/pending jobs per user at the same time.

Figure 4. The "Submit" section in RNArank home page.

Output explanation

The output page of RNArank provides a comprehensive analysis of the input RNA structure's quality. Key components include:

• Input Structure Model: An interactive 3D view of the RNA structure, colored by the predicted per-residue pLDDT score. Regions with higher confidence (blue) are expected to be more accurately modeled.
• pLDDT Plot: A 2D plot showing the per-residue pLDDT score along the sequence, allowing for detailed inspection of local model quality.
• Global pLDDT Score: The global quality score for the entire structure, ranging from 0 to 100. A higher score indicates a better model.

Figure 5. An example of the RNArank result page.

How to cite RNArank?

Please cite the following article when you use the RNArank server:

• Liu et al, Quality assessment of RNA 3D structure models using deep learning and intermediate 2D maps.

Need more help?

If you have more questions or comments about the server, please email yangjysdu.edu.cn.