Part-way through building full
stepwise models, there may be multiple, distinct poses modeling different parts of the overall macromolecule. This score term (see
src/core/scoring/methods/OtherPoseEnergy) is activated on the primary pose, and carries out scoring on poses held in the
other_pose_list. The sums are tallied up into the total score for the primary pose.
Cost (in kBT) of closing loops of uninstantiated residues within pose, or cycles of such loops between pose and other_poses. Assumed a simple Gaussian Chain model, with persistence lengths for RNA/protein based on reasonable guesses. Code for the model are given in
There is an overall offset to this term that corresponds to the cost of confining one end of the loop within Rosetta precision... the value was set based on empirical values of RNA loop modeling energies, but may be way off. Should be possible to compute more rigorously from specialized Monte Carlo calculations.
More info in
src/core/scoring/GaussianChainFunc.cc, including following crazy explanation:
////////////////////////////////////////////////////////////////////////// // --------------------------------- // Closure energies for loop cycles. // --------------------------------- // // D1 // ------~ // ~ | D4 // ~ | // ~ ~ // | ~ // D2 | / // | / D3 // ~~~~~/ // // D1, D2, D3, D4 mark rigid joints, with lengths D1, ... D4. // Squiggles (~~~) mark gaussian chains, // with total variance of gaussian_variance. // // In simplest case (one joint D, gaussian variance sigma^2 ): // // P( closure ) = (capture volume) x ( 1 / 2 pi sigma^2 )^(3/2) // x exp( - D / 2 sigma^2 ) // // [For folks used to thinking about mean end-to-end distance L, // sigma^2 = L^2 / 3 ]. // // Following includes explicit formulae for one, two, three, and four joints. // There is also a general formula which would be straightforward to implement, // which I have worked out (and is basically implemented in the four-joint // function GaussianChainQuadrupleFunc.cc) // // See core/scoring/loop_graph/LoopGraph.cc for use case. // // More notes at: // https://docs.google.com/a/stanford.edu/file/d/0B6gpwdY_Bgd0OHdzVWJGTHBvTzg/edit // // Rhiju, Sep. 2013 // //////////////////////////////////////////////////////////////////////////
Advanced: The energy function above can handle loops that start and end inside a given pose (a 'normal' loop) as well as loops that connect separate poses in the model into a 'cycle'. However, it fails when seeing cycles that are nested into each other, as can happen in the following:
Here D1-A-D2-B-D1 and D1-A-D2-C-D1 could be nested (actually code looks at direction of loops). That means that the penalty for the first cycle (A,B) depends on the influence of C, leading to a complicated integral. Those kinds of pose collections are avoided in
D1 -------- ~ ~ ~ A~ B ~ ~ C ~ ~ ~ -------- D2
stepwiseby checking for "complex loop graphs" in StepWiseMoveSelector.cc. If user specifies
loop_closewill be calculated as a simple sum of loop closure penalties over all cycles -- this is generally an underestimate. In principle, should be possible to approximate the integral by finding the optimal loop configuration given the geometry of connection points (requires numerical solution, but to a convex optimization problem) and computing a log-determinant, but that's not in there yet.
The entropic penalty (in kBT) for each intermolecular connection that is instantiated. Computed as: ( 2.30 - log( concentration / 1 M).
Here the number 2.30 represents log of the effective concentration (relative to 1 M) of one strand relative to another when an interaction is formed. It was calibrated separately based on fits to the nearest-neighbor rules for RNA helix formation; it wil probably be updated later. The assumed reference concentration can be changed from the 'standard' value of 1 M with the flag
-score::conc <Real>; give in units of molar.
Bonuses for virtualizing protein side chains, RNA 5' phosphate, and RNA 2' hydroxyl, respectively. Also stuffing bonuses for virtualizing sugar in
free_suite. These might all get combined into
free_dof for simplicity, after further calibration.
Go back to StepWise Overview.