Rigid body matching/fitting in intermediate-low resolution - PowerPoint Presentation

Slide1

Rigid body matching/fitting in intermediate-low resolution density

Agnel

Praveen

Joseph

STFC, HarwellSlide2

Sample

classigication

Taxonomy

Pfam

domains (fitted models)

Uniprot

ID (text search/Pfam link)

Alignment scores, transformation matrix

Web-server

Database

User’s

query map/ New Entry

User’s

query PDB/

New Entry

Data organisation & annotation

(Work along with the data submission pipeline at EMDB/PDBe and as a development for CCP-EM)Slide3

Methods

D

ensity

based 6D search :

Fast Translational Matching (COLORES)

: Fast Fourier Transform (FFT) driven translational search at each rotation (Chacon and Wriggers, 2002). Cross correlation scores with or without Laplacian filters, are used for matching

Author test cases: Simulated maps down to 34Å resolution were used, and correct alignments were obtained down to 25Å resolution using Laplacian correlation filter. Components of microtubule were also assembled successfully in a 20Å experimental map.Fast Rotational Matching (ADP-EM,)

: Spherical Harmonics accelerated rotational search at each translation (Garzon et al. 2007, Kovacs et al. 2003). Cross correlation scores with or without Laplacian filters, are used for matching

Author test cases: Simulated

maps of down to 30Å were used for testing and accurate results were obtained even at 30Å resolution, using higher harmonic bandwidth values. Fitting components on a 23Å

experimental map was also successful.

Random Sampling (CHIMERA,ModEM):

Randomly sample rotation and translation space (Goddard et al. 2007, Topf et al.2005). Overlap and correlation based scores are

used for matching.

http://3dem.ucsd.edu/SI_Table1_ver2.pdfVilla and Lasker, Finding the right fit: chiseling structures out of cryo

-electron microscopy mapsCurr

Opin Struct Biol. 2014Slide4

Reduced representation :

Gaussian Mixture Model (GMFIT)

: Linear combination of N 3D

gaussian

density (Kawabata, 2008). A gaussian overlap metric is used to score the alignment

.Author test cases:

With the right choice of the number of GDFs, correct alignments were obtained using simulated maps down to 30Å resolution. Tests on a 23.5Å experimental map also gave ‘near-native’ alignment with respect to a reference. FoldEM :

Local density gradients are described as orthogonal feature vectors which are rotationally invariant (represented as vectors covering a set of neighboring grid points: Local Region

Descriptors). Graphs are constructed with these descriptors as nodes and a graph-matching technique is applied to detect the maximal sub-graph. The size of the common sub-graph is returned as the score, along with an atom inclusion score from Chimera (

Pettersen et al., 2004).

Author test cases:

Tests were carried out with simulated EM maps down to 20Å resolution and correct fits were obtained with resolutions down to

15Å. Comparison of 11Å ribosome (experimental) maps in two conformational states also gave a good alignment.

MethodsSlide5

GMfit

Number of 3D

gaussians

used to approximate the density/model depends on the number of local features/components.

Both map and model needs to be converted into

gaussian

mixture models prior to matching

EMD-1056, E-coli 70S, 9Å

EMD-2017, E-coli 30S, 13.5Å

Alignment of gaussian mixturesSlide6

ADP-EM

S

pherical

harmonics are an infinite set of harmonic functions defined on the

sphere.

The basis functions are indexed according to two integer constants, the order, l, and the degree, m. As the

number of coefficients increases, higher frequency signals can be approximated more accurately. The two arguments l and m break the family of polynomials into bands of

functions

Option of

laplacian filter for gradient detectionSlide7

Atomic structure modeling and processing

If

experimentatly

determined structures are not available: Fold recognition and Homology

modeling

Remove flexible loops. Separate

compact domains connected by a hinge/flexible loop.

Check if sub-complexes can be modelled

2631/4umm : 11.6Å

cryo-EM structure of palindromic DNA bound USP/EcR

nuclear receptorMaletta et al. The palindromic DNA-bound USP/EcR

nuclear receptor adopts an asymmetric organization with allosteric domain positioning. Nat comm 2014Slide8

A case: 1534/2y7c/2y7h18Å EcoK1 methyltransferase with a bound

antirestriction

protein

T7 phage

antirestriction

protein

ocrROSETTA(http://robetta.bakerlab.org/) ab-inito model

HsdS-ocr

complex modeled

using guided multi-body docking by HADDOCK (http://haddock.science.uu.nl/services/HADDOCK2.2/haddock.php) HsdM

HsdSSlide9

Volume processing

Shift background peak to zero

Some scores to measure fit quality are sensitive to scale of data values

e.g: overlap and correlation (not about mean) metrics in Chimera use sum of products of density values.

Select a contour threshold

Usually subjective. Can be calculated from molecular weight of sample.Calculations on a subset from EMDB shows a peak at 2 sigmaA higher contour may be considered for local weak density regions:Contour level suggested by authors (EMD-5287, 26Å). Slide10

Search space

Probe/target size ratio should be considered unless you are using an exhaustive search method.

Target map should be segmented to increase probe/target(segment) size ratio.

Gmfit

: Random sampling (

-I R), segmentation (-I SF

) and symmetry based search (-I Y) options are available.adp_em: masking search (-s 2, default): the translational space is limited to positions on which the dimension of the probe (atomic structure) roughly fits inside the experimental EM map. radial search (

-s 1): more uniform and useful for structures with holes.

exhaustive search (-s 0)Chimera

parameter : search N (number of random configurations)Slide11

Translation and rotation Sampling

Depends on resolution/grid spacing – coarse sampling for low resolution maps

Size of probe : finer rotation sampling for a larger probe

adp_em

: higher bandwidth

- finer rotational sampling and more accurate harmonic description. 16 default. Bandwidth values correspond to 360/(2*B) degrees of rotation. translational sampling in Å: -t (one/two voxel steps)Slide12

Try different methods: An example EMD-2631

For intermediate/low

resolution

fitting:

Multiple methods might have to be tried

Different solutions might have to be checked (for those that are structurally stable, functionally relevant and supported by experiments)Multiple scores might be required to re-rank the solutions : surface/envelope matching scores are especially useful at low resolutions.

TUTORIALSlide13

/scratch/ccpem_tutorial/apj_tutorial_22_04/examples/2631module load chimeramodule load adp_emmodule load

gmfit

m

odule load gmconvertSlide14

Volume matching pipeline(ccp-EM/EMDB-PDBe)Slide15

Current version under developmentDatabase of EMDB volume alignments and PDB model – EMDB volume comparisons (fitting)

Web service and

software for volume-volume and model-volume alignments

Web service to search user model/map in EMDB for similar volumes, volumes from certain taxa and sample categories.

Web service and software for 3D analysis of alignments, map feature representations (gaussian

mixtures, points), calculate difference maps.Annotation (sequence/interactions/taxonomy) of entries in EMDBSlide16

Volume pre-processing

Feature points

Contoured density

Gaussian mixture

Dusting

Segmentation

424.8Å

424.8ÅSlide17

Scores (TEMPy)

TEMPy

(Farabella et al.) scores used for re-ranking hits obtained.

Global scores are influenced by non uniform noise, interpolation and padding effects. Local scores used - maps need to be contoured prior to calculations.

If

the resolution of the map is very low to be informative (density variation) – Envelope/

Surface based scores may be useful. Local cross-correlation, Local mutual information, surface distance score and Overlap score are used for re-ranking solutionsSlide18

Examples

E-

coli

70S ribosome:

9

Å vs E-coli 30S ribosome: 13.5Å

Methanococcus maripaludis Mm-cpn chaperone: 4.9Å vs

E-coli

GroEL/GroES

(mut) chaperone: 9.2Å

Human Echovirus 12: 16Å vs

Human

Coxsackievirus A21: 8ÅSlide19

Surface definitions based on a given contour:

Partial surface overlap detection

22.0Å Dengue virus

9

Å 70S E coli vs 6.6Å 80S yeast

Surface feature extraction

Surface point definitionsSlide20

ReleaseTest version of volume matching software is available at: http://www.ccpem.ac.uk/download.php

Currently full functionalities are not included: multiple scores, user map input, more methods to be added (

adp_em

,

shapeEM)

Fully functional version should be released in the next few months.

Web service (EMDB/PDBe) will be also available in the next few months will all functionalitiesSlide21

Martyn Winn

Maya Topf

Ardan Patwardhan

Ingvar Lagerstedt

Thank You