D AMMIF Update Get the latest version of D AMMIF together with the - - PowerPoint PPT Presentation

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

DAMMIF Update

Get the latest version of DAMMIF together with the latest release of ATSAS! ATSAS 2.5.0 will be available soon! http://www.embl-hamburg.de/biosaxs/download.html

Daniel Franke — Ab-Initio Modelling 1/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Ab-Initio Modelling

DAMMIN and DAMMIF Daniel Franke

European Molecular Biology Laboratory

2012/10/19

Daniel Franke — Ab-Initio Modelling 2/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

The following slides describe the how, not the why!

Daniel Franke — Ab-Initio Modelling 3/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Outline

1

Introduction

2

Ab-Initio Modelling

3

Obtaining Models

4

Postprocessing Models

Daniel Franke — Ab-Initio Modelling 4/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Basic Idea

Find a three dimensional object whose theoretical scattering curve would resemble the experimental data best.

Daniel Franke — Ab-Initio Modelling 5/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Results

Daniel Franke — Ab-Initio Modelling 6/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

The Dummy Atom Model

Many little scatterers ...

A Dummy Atom Model (DAM) is build by a tightly packed group of dummy atoms. The volume

• ccupied by dummy atoms in any

state (particle, solvent) is also known as search volume.

Daniel Franke — Ab-Initio Modelling 7/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

The Dummy Atom

One little scatterer ...

Acts as a placeholder for, but does not resemble, a real atom Occupies a known position in space Has a known scattering pattern May either contribute to the solvent or the particle Dummy atoms are also referred to as beads.

Daniel Franke — Ab-Initio Modelling 8/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Basic Idea

Revisited.

Find a three dimensional object whose theoretical scattering curve would resemble the experimental data best. Find the set of dummy atoms within a search volume whose accumulated scattering resembles the experimental data best.

Daniel Franke — Ab-Initio Modelling 9/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Basic Idea

Revisited.

Find a three dimensional object whose theoretical scattering curve would resemble the experimental data best. Find the set of dummy atoms within a search volume whose accumulated scattering resembles the experimental data best.

Daniel Franke — Ab-Initio Modelling 9/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Validity of Input

Garbage In – Garbage Out

Validate input data; check for aggregation at the beginning noise at higher angles Remember: noise can be modelled nicely

Daniel Franke — Ab-Initio Modelling 10/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Outline

1

Introduction

2

Ab-Initio Modelling

3

Obtaining Models

4

Postprocessing Models

Daniel Franke — Ab-Initio Modelling 11/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

An estimate on the problem’s size.

The Universe is not enough

A search volume of 2000 dummy atoms has 22000 ≈ 10600 possible conformations, i.e. scattering curves. On 40.000.000 conformations per hour per CPU, 1000 CPUs, 24 hours a day, 365 days a year one would spend the next couple of universes’ time on enumerating all scattering curves!

Daniel Franke — Ab-Initio Modelling 12/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Imposing restrictions in solution space.

A valid conformation is ... connected: particle beads must be interconnected tightly packed: particle beads shall be tightly packed, avoid loose strands centered: assemble the particle within the search volume, avoid boundary contact in right shape: oblate or prolate shapes can be enforced

Daniel Franke — Ab-Initio Modelling 13/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Advances And Differences In Programs

Selection Scheme DAMMIN DAMMIF

At the current iteration: dark blue particle, might become solvent light blue solvent, might become particle white solvent, won’t change

Daniel Franke — Ab-Initio Modelling 14/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

DAMMIF Walkthrough

$> dammif shape.out

Daniel Franke — Ab-Initio Modelling 15/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

DAMMIF Output

Reading the output of DAMMIF

Step: 1, T: 0.130E-03, 42/1941, Succ: 1229, Eval: 20001, CPU: 00:00:03 Rf: 0.0875, Los: 0.17, Dis: 0.00, Rg: 0.15, Cen:22.57, Ani: 0.00, Fit: 0.0989 Step Step number T Temperature, artifical p/a Number of particle beads of all beads Succ Number of successfull iterations at current T Eval Accumulated number of iterations CPU Accumulated runtime

Daniel Franke — Ab-Initio Modelling 16/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

DAMMIF Output

Reading the output of DAMMIF (cont.)

Step: 1, T: 0.130E-03, 42/1941, Succ: 1229, Eval: 20001, CPU: 00:00:03 Rf: 0.0875, Los: 0.17, Dis: 0.00, Rg: 0.15, Cen:22.57, Ani: 0.00, Fit: 0.0989 Rf Goodness of Fit, data only Los Contribution of Looseness Penalty Dis Contribution of Disconnectivity Penalty Per Contribution of Periphal Penalty Ani Contribution of Anisometry Penalty Fit Goodness of Fit, data and penalties

Daniel Franke — Ab-Initio Modelling 17/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Outline

1

Introduction

2

Ab-Initio Modelling

3

Obtaining Models

4

Postprocessing Models

Daniel Franke — Ab-Initio Modelling 18/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Sequentially on your local machine

Windows; .bat files

dammif lyz.out --mode=slow --prefix FMRP1 dammif lyz.out --mode=slow --prefix FMRP2 dammif lyz.out --mode=slow --prefix FMRP3 dammif lyz.out --mode=slow --prefix FMRP4 dammif lyz.out --mode=slow --prefix FMRP5 dammif lyz.out --mode=slow --prefix FMRP6 dammif lyz.out --mode=slow --prefix FMRP7 dammif lyz.out --mode=slow --prefix FMRP8 dammif lyz.out --mode=slow --prefix FMRP9 dammif lyz.out --mode=slow --prefix FMRP10

Daniel Franke — Ab-Initio Modelling 19/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Sequentially on your local machine

Linux, MacOS; bash syntax

for i in ‘seq 1 10‘ ; do dammif --prefix=lyz-\$i --mode=slow lyz.out; done

Daniel Franke — Ab-Initio Modelling 20/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

In parallel on your local cluster

Please contact your system administrator for details of your cluster and how to submit jobs. Important: as processes are being run in parallel, multiple may be started at the same time – with the same random seed – resulting in exactly the same model. Make sure to redefine the random seed for each run!

Daniel Franke — Ab-Initio Modelling 21/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Input redirection

Fine tuning parameters in scripts

1 Start dammif in slow mode once, abort 2 Find the $prefix.in file 3 Modify as needed 4 Run dammif as

$> dammif --prefix=... --mode=i < modified.in

Daniel Franke — Ab-Initio Modelling 22/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

In parallel using ATSAS-Online

http://www.embl-hamburg.de/biosaxs/atsas-online/

Create an account (email address only) and submit your dammin/dammif jobs to the EMBL BioSAXS cluster.

Daniel Franke — Ab-Initio Modelling 23/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

In parallel on the GRID

http://www.wenmr.org/wenmr/ab-initio-modelling

A worldwide e-Infrastructure for NMR and structural biology. In preparation and not yet available.

Daniel Franke — Ab-Initio Modelling 24/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Outline

1

Introduction

2

Ab-Initio Modelling

3

Obtaining Models

4

Postprocessing Models

Daniel Franke — Ab-Initio Modelling 25/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Postprocessing Models

How to proceed ...

With multiple models: find those that are most similar (uniqueness of reconstruction is not guaranteed) OR group models into clusters superimpose and average the selection restart fitting process using the averaged model

Daniel Franke — Ab-Initio Modelling 26/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Multiple models

Funari et al. (2000) J. Biol. Chem. 275, 31283–31288.

5S RNA, multiple solutions with equally good fit.

Daniel Franke — Ab-Initio Modelling 27/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Selecting Models

DAMSEL, DAMCLUST

Computes the similarities between all pairs of input files. Measure of similarity of models: Normalized Spatial Discrepancy (NSD) NSD < 1 implies similar models

DAMSEL selects similar models, rejects outliers DAMCLUST groups models to clusters, rejects nothing

Daniel Franke — Ab-Initio Modelling 28/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Superimposing Models

SUPCOMB, DAMSUP

SUPCOMB: superimpose any two models

(principle axis alignment, gradient minimization, local grid search)

DAMSUP: superimpose multiple models on a

reference using SUPCOMB.

Daniel Franke — Ab-Initio Modelling 29/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Superimposing models

5S RNA continued ...

Solution spread region.

Daniel Franke — Ab-Initio Modelling 30/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Superimposing models

5S RNA continued ...

Solution spread region. Most populated volume.

Daniel Franke — Ab-Initio Modelling 30/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Averaging Models

DAMAVER, DAMFILT

DAMAVER: Creates a bead probability density map

within the search volume.

DAMFILT: Generates the averaged model, using a

user-defined probability threshold. Will give a valid model, violating the threshold if necessary.

Daniel Franke — Ab-Initio Modelling 31/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Ab-Initio Modelling

Options at this point.

take the model(s) that have the least NSD to all

• thers – this fits the data

take the filtered model(s) – but this will not fit the data use averaged model(s) and restart DAMMIN to fit the experimental data (via DAMSTART)

Daniel Franke — Ab-Initio Modelling 32/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Ab-Initio Modelling

Options at this point.

take the model(s) that have the least NSD to all

• thers – this fits the data

take the filtered model(s) – but this will not fit the data use averaged model(s) and restart DAMMIN to fit the experimental data (via DAMSTART)

Daniel Franke — Ab-Initio Modelling 32/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Ab-Initio Modelling

Options at this point.

take the model(s) that have the least NSD to all

• thers – this fits the data

take the filtered model(s) – but this will not fit the data use averaged model(s) and restart DAMMIN to fit the experimental data (via DAMSTART)

Daniel Franke — Ab-Initio Modelling 32/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Postprocessing Models

Summary.

Daniel Franke — Ab-Initio Modelling 33/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

Ab-Initio Modelling

5S RNA continued ...

Finalized model, filtered by DAMSTART, refined by

DAMMIN.

Daniel Franke — Ab-Initio Modelling 34/35

Introduction Ab-Initio Modelling Obtaining Models Postprocessing Models

That’s all folks.

Questions? Visit http://www.saxier.org/forum

Daniel Franke — Ab-Initio Modelling 35/35