k 2016

Direct SAXS Curve Fitting with an Ensable of Conformations

KŘENEK, Aleš, Tomáš RAČEK a Jana PAZÚRIKOVÁ

Základní údaje

Originální název

Direct SAXS Curve Fitting with an Ensable of Conformations

Název česky

Přímé fitování křivek SAXS na ansámbl konformací

Název anglicky

Direct SAXS Curve Fitting with an Ensable of Conformations

Vydání

Molecular Machines 2016, 2016

Další údaje

Typ výsledku

Prezentace na konferencích

Utajení

není předmětem státního či obchodního tajemství

Klíčová slova anglicky

SAXS; ensamble fit
Změněno: 23. 8. 2017 13:18, Mgr. Aleš Křenek, Ph.D.

Anotace

V originále

Fitting a computed SAXS scattering curve onto its experimental counterpart is an essential task in processing the SAXS data, implemented in several toolkits (Svergun 1995, Schneidman-Duhovny 2013). Quality of the fit can be used to choose from among several candidate conformations. The process gets more complicated when an unknown mixture of the candidates is expected. Methods available so far (Pelikan 2009, Tria 2015) use genetic algorithms to determine a small subset of the ensamble, followed by a search of the weights. We propose an alternate direct method based on randomized global optimization. The search space is formed by weights of all the candidate conformations and two more parameters (characteristics of solvent and solvent accessible surface in Debye formula). The global optimization combines either stochastic tunelling or random walk with constrained local optimization (Powell 2009). The two appoaches give comparable results -- random walk is likely to hit a good fit faster but it must polish the solution with local optimization in every step, while more systematic stochastic tunelling proceeds with shorter but faster steps. It appears that the optimization picks the right subset of candidates seemlessly, quickly eliminating the others by setting their weights to zero. The method is limited by the number of dimensions of the search space, i.e. the number of candidate conformations, several hundreds are applicable. Therefore it is suitable for for combined NMR+SAXS experiments; NMR restraints are processed first (e.g. with Cyana) to yield a few dozens of candidates, and combined with SAXS using our tool then. Practical computation takes few minutes with current CPUs typically; its complexity is quadratic w.r.t. number of atoms due to straightforward implementation of the Debye formula. We present example results with few moderate-size proteins. The application is available to the community also as a web portal at http://saxs.cerit-sc.cz.

Anglicky

Fitting a computed SAXS scattering curve onto its experimental counterpart is an essential task in processing the SAXS data, implemented in several toolkits (Svergun 1995, Schneidman-Duhovny 2013). Quality of the fit can be used to choose from among several candidate conformations. The process gets more complicated when an unknown mixture of the candidates is expected. Methods available so far (Pelikan 2009, Tria 2015) use genetic algorithms to determine a small subset of the ensamble, followed by a search of the weights. We propose an alternate direct method based on randomized global optimization. The search space is formed by weights of all the candidate conformations and two more parameters (characteristics of solvent and solvent accessible surface in Debye formula). The global optimization combines either stochastic tunelling or random walk with constrained local optimization (Powell 2009). The two appoaches give comparable results -- random walk is likely to hit a good fit faster but it must polish the solution with local optimization in every step, while more systematic stochastic tunelling proceeds with shorter but faster steps. It appears that the optimization picks the right subset of candidates seemlessly, quickly eliminating the others by setting their weights to zero. The method is limited by the number of dimensions of the search space, i.e. the number of candidate conformations, several hundreds are applicable. Therefore it is suitable for for combined NMR+SAXS experiments; NMR restraints are processed first (e.g. with Cyana) to yield a few dozens of candidates, and combined with SAXS using our tool then. Practical computation takes few minutes with current CPUs typically; its complexity is quadratic w.r.t. number of atoms due to straightforward implementation of the Debye formula. We present example results with few moderate-size proteins. The application is available to the community also as a web portal at http://saxs.cerit-sc.cz.