An overview of this energy function optimization effort is available here.
For information on the previous set of updates to this score function, see the beta nov15 score function page here.
For most protocols (those that use getScoreFunction to set the protocol score function), the flag -beta_nov16 or simply -beta (which will always load the latest beta energy function) will load this version of the beta energy function.
For RosettaScripts protocols, the flag -beta_nov16 or -beta must be provided, and the following scorefunction declaration must be made:
<beta weights=beta_nov16/> or <beta weights=beta/>
Optimization followed the same scheme as beta_nov15, with new decoy sets for docking generated using beta_nov16. Optimized parameters include:
Several rounds of refinement were carried out in which subsets of these terms were considered.
Bridging water potential (lk_bridge and lk_bridge_uncpl)
A term assessing the contribution of bridging water molecules has been incorporated into this scorefunction variant. The term places ideal virtual waters on all polar groups, and the overlap between these virtual waters is provided as a bonus (the fade as the overlap decreases is fit as a parameter). Additionally, the angle between the base atoms controls the strength of this bonus. It is broken into two subterms:
In beta_nov16, both terms are given equal weight, approximately -0.3.
Intra-etable energies and torsional corrections
In beta_nov16, all intra-res etable energies and intra-res fa_elec are enabled at the same weight as the corresponding intra-residue term. All use Rhiju's xover4 variants to make count pair behavior consistent both intra- and inter-residue. To avoid double counting, the torsional correction scheme of Conway and DiMaio (Prot. Sci. 2016) is applied to both fa_dun and rama.
Enabling these intra-res energies should (hopefully) improve interoperability with non-canonic amino acids as well as ligands/nucleic acids.
sp3 hbond acceptors
A modified potential for sp3 hbond acceptors has been implemented. Rather than use an explicit torsional potential, it instead uses a "soft-max" potential between both "base atoms" (e.g. for serine, the angular term becomes soft-max(f_ang(CB-OG-Hdon),f_ang(HG-OG-Hdon) ). This has been enabled by default for water since the previous potential was incorrect. It is enabled for all sp3 acceptors in -beta_nov16.
Reference weight fitting
A new scheme for reference-weight fitting was used, where -- following fitting of all other terms -- we optimized the weighted sum of the following:
For the latter two metrics, rank correlation between experiment and simulation was used as the target function. Weights on each set were chosen so that no individual test worsened significantly (>1% recovery or >0.01 correlation) during fitting.
Other minor changes