Loop modeling method

The loop modeling method first takes the generated model, and selects all standard residues around gaps in the alignment for additional loop modeling. (To select a different region for modeling, simply redefine the LoopModel.select_loop_atoms() routine to select the relevant atoms.) An initial loop conformation is then generated by simply positioning the atoms of the loop with uniform spacing on the line that connects the main-chain carbonyl oxygen and amide nitrogen atoms of the N- and C-terminal anchor regions respectively (to change this, override the LoopModel.build_ini_loop() method), and this model is written out to a file with the .IL extension.

Next, a number of loop models are generated from LoopModel.loop.starting_model to LoopModel.loop.ending_model. Each takes the initial loop conformation and randomizes it by ± 5Å in each of the Cartesian directions. The model is then optimized thoroughly twice, firstly considering only the loop atoms and secondly with these atoms “feeling” the rest of the system. The loop optimization relies on an atomistic distance-dependent statistical potential of mean force for nonbond interactions [Melo & Feytmans, 1997]. This classifies all amino acid atoms into one of 40 atom classes (as defined in $LIB/atmcls-melo.lib) and applies a potential as MODELLER cubic spline restraints (as defined in $LIB/melo-dist1.lib). No homology-derived restraints are used during this procedure. Each loop model is written out with the .BL extension.

For more information, please consult the loop modeling paper [Fiser et al., 2000] or look at the loop modeling class itself in modlib/modeller/automodel/loopmodel.py.