Is Mfold supervised?

1 reply [Last post]
Joined: 03/24/2011


I'm trying to improve the Do's Contrafold( by semi-supervised learning. While he used RNA sequences with given reliable structures as training data, we also involve a large amount of RNA sequences without known structures to improve the performance.

He cited your Mfold in his paper, made a comparison.
So I have a brief question. Is Mfold supervised, as same as Contrafold, which means the training data must be a pair of sequence and its known structure? Or as semi-supervised learning, the raw sequence can also be used in Mfold in training process?


Ming Tan

Joined: 11/12/2010
"supervised" learning: mfold versus contrafold

The mfold software is most definitely not supervised. In fact, I don't even understand what the term means in this context. The same is true for the UNAFold software. Both of these packages use deterministic algorithms together with free energy parameters deduced from melting experiments. The software is in no way "trained" by a collection of sequences whose structures are known a priori. I hope that helps.