Experimental Structure Probing Data#
While RNA secondary structure prediction yields good predictions in general, the model implemented in the prediction algorithms and its parameters are not perfect. This may be due to several reasons, such as uncertainties in the parameters and the simplified assumptions of the model itself. However, prediction performance can be increased by integrating (experimental) RNA structure probing data, such as derived from selective 2’-hydroxyl acylation analyzed by primer extension (SHAPE), dimethyl sulfate (DMS), inline probing, or similar techniques.
Such experimental probing data is usually integrated in the form of small pertubations in the evaluated energy contributions (Soft Constraints) that effectively guide the prediction towards the information gained from the experiment.
Structure Probing Data Workflow#
Our implementations follow a particular workflow that basically divides the application of the structure probing data into three steps:
Pre-processing of the probing data
Conversion of the pre-processed data into pseudo energy contributions
Passing the pseudo energies to the soft-constraints interface
The first two steps are handled by a dedicated strategy, that basically is a
single callback function (vrna_probing_strategy_f). Then, the probing data and
the strategy need to be bundled, e.g. by vrna_probing_data_linear(), to obtain
a data structure of type vrna_probing_data_t. This data structure is then passed
to the vrna_sc_probing() function that performs the last of the three steps by
calling the strategy to process the data and adding the derived pseudo energy
contributions as soft constraints.
We already implemented a few different strategies that can be used out-of-the-box, i.e. we provide wrappers that generate the bundled data structure for a particular strategy. The available API symbols can be found in probing/strategies.
However, the generic interface allows for any user-defined strategy, thus new strategies can be easily implemented by the user and then passed to our API.
After all the above three steps are done, any prediction (MFE, partition function, etc.) will acknowledge the probing data and therefore guide the prediction based on it.
In the following, you’ll find the respective API symbols that allow for the integration of experimental probing data.
Strategies for Linear Data#
Recent literature lists several methods and strategies that deal with the problem to convert linear experimental RNA structure probing data into pseudo energy terms to guide RNA secondary strucure predictions. Such data may come from SHAPE, DMS, lead, inline probing, or similar techniques. The most commonly used method may be the one presented in Deigan et al. [2009] where the free energy evaluation for stacked base pairs adds a pseudo energy contribution derived from the reactivity values of the probing data. The higher the reactivity the stronger the stack is penalized by a positive energy.
SHAPE Reactivity Data#
Incorporate SHAPE reactivity structure probing data into the folding recursions by means of soft constraints.
Details for our implementation to incorporate SHAPE reactivity data to guide secondary structure prediction can be found in Lorenz et al. [2016].
Generate Soft Constraints from Data#
Find a vector of perturbation energies that minimizes the discripancies between predicted and observed pairing probabilities and the amount of neccessary adjustments.