[Question] ThreshKnot implementation used in LinearTurboFold

[waltergallegog]

Hello, I have a simple question regarding the different ThreshKnot implementations available:

• In the LTF paper, you mention as reference for ThreshKnot the following paper: L. Zhang, H. Zhang, D. H. Mathews, L. Huang, ThreshKnot: Thresholded prob-knot for improved RNA secondary structure prediction. arXiv [Preprint] (2019). In this paper, it says: ThreshKnot is available in the RNAstructure software package v6.2 (released November 27, 2019):https://rna.urmc.rochester.edu/RNAstructure.html

• The LinearPartition package offers the option --threshknot or -T to get the threshknot structure

• In LinearTurboFold however, it seems to me threshknot has been implemented again, here

Could you confirm if those 3 implementations are equivalent, or has something been changed in the LTF implementation that could lead to better structures?

Thanks again Walter.

[sizhen]

Hey Walter,

LinearPartition and LinearTurboFold incorporate ThreshKnot to make sure the end-to-end runtime is linear with the sequence length. And yes, they are equivalent. The implementation in LinearTurboFold is the same as the one in RNAstructure.

The prediction would be the same as long as the parameters of ThreshKnot are the same (threshold, the minimum length of the helix, and iterations). And the major parameter is the value of the threshold, the other two parameters wouldn't change the accuracy too much.

Sizhen

[waltergallegog]

Thanks a lot for the clarification. BR Walter