Supporting Material for the Paper: Rich Parameterization Improves RNA Structure Prediction

We specify here the features which are used by our models. Each feature description is composed of two parts: a description of a structural element, and a (possibly empty) description of a sequential context. All models discussed in the paper are obtained by combining a set of structural elements St with a set of sequential contexts Co, and producing all corresponding features (i.e. producing a feature for each structural element in St and a corresponding sequential context from Co). In Section 1 we define the different loop types, which are used for categorizing structural elements. In Section 2 we give the three sets of structural elements St, St, and St, and in Section 3 we give the three sets of sequential contexts Co, Co, and Co, which are used in our models. All examples in the text refer to the sequence-folding (x, y) depicted in Fig. 1.