Sigmoidal crosslinks in IMP.pmi 2.7.0
I am using IMP.pmi to model a protein complex using MS crosslinks. I notice that the IMP.pmi crosslinking function, IMP.pmi.restraints.crosslinking.CrossLinkingMassSpectrometryRestraint(), scores crosslinks basically as a linear function of crosslink distance. The crosslink length is basically irrelevant here, shorter=better.
Ideally, I would like to score crosslinks as a sigmoid function, with crosslinks inside the specified 'length' dropping in score. I notice that there is a deprecated SigmoidCrossLinkMS() function in IMP, which uses an IMP.pmi.SigmoidRestraintSphere to model the crosslinks. Is there a reason this is deprecated? Should I be using it? It seems to offer functionality not available elsewhere.
Regards, Fergal
Fergal Duffy Postdoctoral Researcher Center for Infectious Disease Researchhttp://www.cidresearch.org/ 307 Westlake Ave N, Suite 500 | Seattle, WA 98109 206.256.7435 | fergal.duffy@cidresearch.org
On 6/8/17 4:41 PM, Fergal Duffy wrote: > I am using IMP.pmi to model a protein complex using MS crosslinks. I > notice that the IMP.pmi crosslinking function, > IMP.pmi.restraints.crosslinking.CrossLinkingMassSpectrometryRestraint(), > scores crosslinks basically as a linear function of crosslink distance. > The crosslink length is basically irrelevant here, shorter=better.
I don't think that's right, although Riccardo would know for sure. The restraint *does* include a linear term, but my understanding is that's intended to gradually pull in residue pairs that far exceed the crosslink length. At shorter distances the scoring is Bayesian, and certainly does use the crosslink length.
> Ideally, I would like to score crosslinks as a sigmoid function, with > crosslinks inside the specified ‘length’ dropping in score. I notice > that there is a deprecated SigmoidCrossLinkMS() function in IMP, which > uses an IMP.pmi.SigmoidRestraintSphere to model the crosslinks. Is there > a reason this is deprecated? Should I be using it? It seems to offer > functionality not available elsewhere.
It's deprecated because the current Bayesian formalism is superior, but there are some older published works that used that restraint (and we want them to still be reproducible). It's probably not a good idea to use it in modern applications.
Ben
IMP.pmi.restraints.crosslinking.CrossLinkingMassSpectrometryRestraint()
is indeed a sigmoidal function modulated by two parameters sigma (inflection) and psi (baselines). The maximal length of the crosslinker is passed through "length=..." in the constructor. There is a linear part, but that is just a soft bias to avoid particles getting dispersed, since the potential is flat at high distance values. The slope of the linear part is 0.01-0.02 Score Unit/ Angstrom, can be changed using "slope=..."
Riccardo Pellarin, PhD =================== Institut Pasteur CNRS UMR 3528 25, rue du Docteur Roux 75724 Paris Cedex 15, France riccardo.pellarin@pasteur.fr +33 (0)1 44 38 93 63
On Fri, Jun 9, 2017 at 1:41 AM, Fergal Duffy fergal.duffy@cidresearch.org wrote:
> I am using IMP.pmi to model a protein complex using MS crosslinks. I > notice that the IMP.pmi crosslinking function, IMP.pmi.restraints. > crosslinking.CrossLinkingMassSpectrometryRestraint(), scores crosslinks > basically as a linear function of crosslink distance. The crosslink length > is basically irrelevant here, shorter=better. > > > > Ideally, I would like to score crosslinks as a sigmoid function, with > crosslinks inside the specified ‘length’ dropping in score. I notice that > there is a deprecated SigmoidCrossLinkMS() function in IMP, which uses an > IMP.pmi.SigmoidRestraintSphere to model the crosslinks. Is there a reason > this is deprecated? Should I be using it? It seems to offer functionality > not available elsewhere. > > > > Regards, > > Fergal > > > > > > > > > *Fergal Duffy *Postdoctoral Researcher > > Center for Infectious Disease Research http://www.cidresearch.org/ > 307 Westlake Ave N, Suite 500 | Seattle, WA 98109 > 206.256.7435 <%28206%29%20256-7435> | *fergal.duffy@cidresearch.org > fergal.duffy@cidresearch.org* > > > > _______________________________________________ > IMP-users mailing list > IMP-users@salilab.org > https://salilab.org/mailman/listinfo/imp-users > >
I would also like to know why the sigmoid scoring function was deprecated.
thanks,
Isabelle
On Jun 29, 2017, at 2:34 AM, Riccardo Pellarin <pellarin.riccardo@gmail.commailto:pellarin.riccardo@gmail.com> wrote:
IMP.pmi.restraints.crosslinking.CrossLinkingMassSpectrometryRestraint()
is indeed a sigmoidal function modulated by two parameters sigma (inflection) and psi (baselines). The maximal length of the crosslinker is passed through "length=..." in the constructor. There is a linear part, but that is just a soft bias to avoid particles getting dispersed, since the potential is flat at high distance values. The slope of the linear part is 0.01-0.02 Score Unit/ Angstrom, can be changed using "slope=..."
Riccardo Pellarin, PhD =================== Institut Pasteur CNRS UMR 3528 25, rue du Docteur Roux 75724 Paris Cedex 15, France riccardo.pellarin@pasteur.frmailto:riccardo.pellarin@pasteur.fr +33 (0)1 44 38 93 63
On Fri, Jun 9, 2017 at 1:41 AM, Fergal Duffy <fergal.duffy@cidresearch.orgmailto:fergal.duffy@cidresearch.org> wrote: I am using IMP.pmi to model a protein complex using MS crosslinks. I notice that the IMP.pmi crosslinking function, IMP.pmi.restraints.crosslinking.CrossLinkingMassSpectrometryRestraint(), scores crosslinks basically as a linear function of crosslink distance. The crosslink length is basically irrelevant here, shorter=better.
Ideally, I would like to score crosslinks as a sigmoid function, with crosslinks inside the specified ‘length’ dropping in score. I notice that there is a deprecated SigmoidCrossLinkMS() function in IMP, which uses an IMP.pmi.SigmoidRestraintSphere to model the crosslinks. Is there a reason this is deprecated? Should I be using it? It seems to offer functionality not available elsewhere.
Regards, Fergal
Fergal Duffy Postdoctoral Researcher Center for Infectious Disease Researchhttp://www.cidresearch.org/ 307 Westlake Ave N, Suite 500 | Seattle, WA 98109 206.256.7435tel:%28206%29%20256-7435 | fergal.duffy@cidresearch.orgmailto:fergal.duffy@cidresearch.org
_______________________________________________ IMP-users mailing list IMP-users@salilab.orgmailto:IMP-users@salilab.org https://salilab.org/mailman/listinfo/imp-users
_______________________________________________ IMP-users mailing list IMP-users@salilab.orgmailto:IMP-users@salilab.org https://salilab.org/mailman/listinfo/imp-users
Looking at the IMP.pmi.restraints.crosslinking.CrossLinkingMassSpectrometryRestraint() source, the only parameters I can see are slope and length, there does not appear to be any sigma parameter to set an inflection point where the exponential portion of the function tops out.
The CrossLinkingMassSpectrometryRestraint score gets evaluated using an IMP.isd.CrosslinkMSRestraint(), where the score is calculated at IMP.isd.CrossLinkMSRestraint.cpp at line 149 as -log(get_probability()), where the probability is given a prior of exp(-slope_*dist) at line 100. Taking a log of an exponent seems to indicate that the score is an linear function of the distance, where the crosslink length is only taken into account indirectly via the psi and pofr variables.
When I set the slope parameter as any value much above 0.025, IMP cannot start the simulation, as it results in a frame 0 score Inf in the simulation output, due to the unbounded exponential increase in score for far-apart randomly shuffled crosslinks. Using the SigmoidCrossLinkMS to model crosslinks does work as expected, with an intial S-shaped step increase in score followed by a linear increase. I had to modify the SigmoidCrossLinkMS() class code to take a Hierarchy object instead of a Representation, but it seems to behave much more nicely.
Fergal Duffy Postdoctoral Researcher Center for Infectious Disease Researchhttp://www.cidresearch.org/ 307 Westlake Ave N, Suite 500 | Seattle, WA 98109 206.256.7435 | fergal.duffy@cidresearch.org
From: imp-users-bounces@salilab.org [mailto:imp-users-bounces@salilab.org] On Behalf Of Riccardo Pellarin Sent: Thursday, June 29, 2017 2:34 AM To: Help and discussion for users of IMP Subject: Re: [IMP-users] Sigmoidal crosslinks in IMP.pmi 2.7.0
IMP.pmi.restraints.crosslinking.CrossLinkingMassSpectrometryRestraint() is indeed a sigmoidal function modulated by two parameters sigma (inflection) and psi (baselines). The maximal length of the crosslinker is passed through "length=..." in the constructor. There is a linear part, but that is just a soft bias to avoid particles getting dispersed, since the potential is flat at high distance values. The slope of the linear part is 0.01-0.02 Score Unit/ Angstrom, can be changed using "slope=..."
Riccardo Pellarin, PhD =================== Institut Pasteur CNRS UMR 3528 25, rue du Docteur Roux 75724 Paris Cedex 15, France riccardo.pellarin@pasteur.frmailto:riccardo.pellarin@pasteur.fr +33 (0)1 44 38 93 63
On Fri, Jun 9, 2017 at 1:41 AM, Fergal Duffy <fergal.duffy@cidresearch.orgmailto:fergal.duffy@cidresearch.org> wrote: I am using IMP.pmi to model a protein complex using MS crosslinks. I notice that the IMP.pmi crosslinking function, IMP.pmi.restraints.crosslinking.CrossLinkingMassSpectrometryRestraint(), scores crosslinks basically as a linear function of crosslink distance. The crosslink length is basically irrelevant here, shorter=better.
Ideally, I would like to score crosslinks as a sigmoid function, with crosslinks inside the specified ‘length’ dropping in score. I notice that there is a deprecated SigmoidCrossLinkMS() function in IMP, which uses an IMP.pmi.SigmoidRestraintSphere to model the crosslinks. Is there a reason this is deprecated? Should I be using it? It seems to offer functionality not available elsewhere.
Regards, Fergal
Fergal Duffy Postdoctoral Researcher Center for Infectious Disease Researchhttp://www.cidresearch.org/ 307 Westlake Ave N, Suite 500 | Seattle, WA 98109 206.256.7435tel:%28206%29%20256-7435 | fergal.duffy@cidresearch.orgmailto:fergal.duffy@cidresearch.org
_______________________________________________ IMP-users mailing list IMP-users@salilab.orgmailto:IMP-users@salilab.org https://salilab.org/mailman/listinfo/imp-users
participants (4)
-
Ben Webb -
Fergal Duffy -
Isabelle Phan -
Riccardo Pellarin