Merge pull request #3131 from CyrusBiotechnology/smlewis/rescue_indigo/nmer_refactor_plus_commonsmerge2
This PR merges @indigogo 's re-factored NMerSVMEnergy (currently used by Cyrus) into the Rosetta master branch, and now uses the `EnergyMethodOptions` system (courtesy of @vmullig ). This paves the way for @cbaileykellogg and myself to complete and merge #3464 (`MHCEpitopeEnergy`), which integrates nmer into the packer-compatible mhc_energy scoreterm.
Integration test failures are due to writing out `EnergyMethodOptions` lists to PDB files, and are purely cosmetic.
@indigogo 's description of these changes:
Changes to NMerSVMEnergy scoring method and associated filter. This adds normalized score rank percentage calculation to NMerSVMEnergy. Why? Here's the problem: when using NMerSVMEnergy to predict T cell epitopes from MHC binding predictions, some MHC allele types bind more or less strongly on average. However, this is not indicative of more or less average predisposition to causing T cell activation. To correctly combine predictions across a repertoire of MHC types, we need to normalize the scores reported by different SVM models. To do this, I predicted MHC binding for 100k random 15mers from the human genome for every allele svm in the rosetta database and stored those distributions back in the database. Every time NMerSVMEnergy calculates a score, it also calculates the rank fraction of that score against the precalculated distribution; this info is retrievable from the scoring method directly, or you can pass a non-default cmdline option to switch to reporting that rank fraction as the score terms 'energy' instead of using the raw score. This allows one to properly combine the predictions of multiple alleles during deimmunization design in rosetta.
Also tagging @smlewis ,