Hello IMPers,
I am continuing to working with IMP/PMI for integrative modeling of a protein complex using EM and crosslinking information. Ultimately, as in the tutorial, I have 4 sets of restraints: excluded volume, connectivity, crosslinks and gaussian.
I would like to explore what weights to set. At the moment, I am doing a grid search and looking at within-simulation and between-simulation RMSD of the best models. Perhaps you can point me towards some resources or guidelines on how to optimise weights in IMP? I know this is a big topic and a big field, is there perhaps some optimisation script or some sort of wrapper you can point me towards?
In general, it seems to me that in the tutorial XL, EV and connectivity restraints are left with the default weights, and the Gaussian restraint is scaled so that it has similar violation scored as the XL restraints. Is this a sort of "recommended starting point", provided I have equal confidence in XL and EM data?
Additionally, I would love to implement some sort of multi-stage sampling where particles are initially fit to the EM density with low XL weights, and later refined using a higher XL weight. Is it possible to somehow pass the best energy model of one ReplicaExchange macro as the starting point of a second one?
Many thanks for your input,
Andrea Graziadei PostDoc, Rappsilber Group Tu Berlin
On 4/26/19 3:40 AM, Andrea Graziadei wrote: > I am continuing to working with IMP/PMI for integrative modeling of a > protein complex using EM and crosslinking information. Ultimately, as in > the tutorial, I have 4 sets of restraints: excluded volume, > connectivity, crosslinks and gaussian. > > I would like to explore what weights to set.
Generally it shouldn't be necessary to modify the weights, except for possibly the EM restraint. There thus isn't really a protocol for optimizing the weights. Ideally each restraint would be Bayesian in nature such that weights are entirely unnecessary, although this isn't yet the case for EM.
Ben