Structure databases, searches and alignments Marian Novotny [email protected] Molecular Bioinformatics X3.

Structure databases, searches and alignments

Marian Novotny

[email protected] Molecular Bioinformatics X3

Outline

1. Structure databases - why do we need them?

- types of structural databases

- Protein Data Bank

- other useful databases

2. Searches - text searches

3. Structure searches (alignments) - why?

- how ?

- comparison of available tools

Structure databases

Why?

data tend to get lost

source of information for further analysis

better access to data by general public

validation of data is (sometimes) possible

…. a structured collection of data held in computer storage; esp. one that incorporates software to make it accessible in a variety of ways; transf., any large collection of information.

Oxford English dictionary

Database is…

…..a usually large collection of data organized especially for rapid search and retrieval (as by a computer)

Merriam-Webster Online

Databases

Primary databasesAdded-value databases Derived databases

RCSB

MSD

PDBJ

NDB

CSD

OCA

PDBSum EDS

Whatcheck

Jena Image library

ftp archive of flat files

Primary databases

- repositories of experimental data of macromolecular structures (X-ray, NMR, electron microscopy…)

- RCSB (USA), MSD (Europe) and PDBJ (Japan) collaborate to form wwPDB. Data can be submitted to any of these databases. Databases interchange their new data on a regular basis, so they have an identical content.

- primary databases differ in presentation of data and the amount of extra services and links they provide

The Protein Data Bank (PDB)

- established in 1971 by Walter Hamilton at Brookhaven National Laboratory

- seven structures were deposited at the beginning

- the database was distributed on magnetic tapes

- RCSB now run by the consortium of three institutions (San Diego Supercomputer Centre, Rutgers University and Centre for Avanced Reasearch and Biotechnology)

- 29326 structures (26.01.2005)

- distributed over internet

- released once a week

HEADER HYDROLASE 27-OCT-03 1UR9 TITLE INTERACTIONS OF A FAMILY 18 CHITINASE WITH THE DESIGNED TITLE 2 INHIBITOR HM508, AND ITS DEGRADATION PRODUCT, TITLE 3 CHITOBIONO-DELTA-LACTONE COMPND MOL_ID: 1; COMPND 2 MOLECULE: CHITINASE B; COMPND 3 CHAIN: A, B; COMPND 4 EC: 3.2.1.14; COMPND 5 ENGINEERED: YES; COMPND 6 MUTATION: YES SOURCE MOL_ID: 1; SOURCE 2 ORGANISM_SCIENTIFIC: SERRATIA MARCESCENS; SOURCE 3 STRAIN: BJL200; SOURCE 4 EXPRESSION_SYSTEM: ESCHERICHIA COLI; SOURCE 5 EXPRESSION_SYSTEM_STRAIN: DH5 ALPHA; SOURCE 6 OTHER_DETAILS: CLONED GENE KEYWDS CHITINASE, INHIBITION, LACTONE, CHITIN DEGRADATION, KEYWDS 2 HYDROLASE, GLYCOSIDASE EXPDTA X-RAY DIFFRACTION AUTHOR G.VAAJE-KOLSTAD,A.VASELLA,M.G.PETER,C.NETTER,D.R.HOUSTON, AUTHOR 2 B.WESTERENG,B.SYNSTAD,V.G.H.EIJSINK,D.M.F.VAN AALTEN REVDAT 1 27-APR-04 1UR9 0 JRNL AUTH G.VAAJE-KOLSTAD,A.VASELLA,M.G.PETER,C.NETTER, JRNL AUTH 2 D.R.HOUSTON,B.WESTERENG,B.SYNSTAD,V.G.H.EIJSINK JRNL AUTH 2 D.M.F.VAN AALTEN JRNL TITL INTERACTIONS OF A FAMILY 18 CHITINASE WITH THE JRNL TITL 2 DESIGNED INHIBITOR HM508 AND ITS DEGRADATION JRNL TITL 3 PRODUCT, CHITOBIONO-DELTA-LACTONE JRNL REF J.BIOL.CHEM. V. 279 3612 2004 JRNL REFN ASTM JBCHA3 US ISSN 0021-9258 REMARK 1 REMARK 1 REFERENCE 1 REMARK 1 AUTH D.M.F.VAN AALTEN,D.KOMANDER,B.SYNSTAD,S.GASEIDNES, REMARK 1 AUTH 2 M.G.PETER,V.G.H.EIJSINK REMARK 1 TITL STRUCTURAL INSIGHTS INTO THE CATALYTIC MECHANSIM OF REMARK 1 TITL 2 A FAMILY 18 EXOCHITINASE REMARK 1 REF PROC.NAT.ACAD.SCI.USA V. 98 8979 2001 REMARK 1 REFN ASTM PNASA6 US ISSN 0027-8424 REMARK 1 REFERENCE 2 REMARK 1 AUTH D.M.F.VAN AALTEN,B.SYNSTAD,M.B.BRURBERG,E.HOUGH, REMARK 1 AUTH 2 B.RIISE,V.G.H.EIJSINK,R.K.WIERENGA REMARK 1 TITL STRUCTURE OF A TWO-DOMAIN CHITOTRIOSIDASE FROM

PDB FILE

12345678901234567890123456789012345678901234567890123456789012345678901234567890 1 2 3 4 5 6 7 8ATOM 340 N PHE A 43 3.853 28.346 32.161 1.00 10.57 N ATOM 341 CA PHE A 43 3.839 29.688 32.724 1.00 12.33 C ATOM 342 C PHE A 43 3.096 29.747 34.047 1.00 13.20 C ATOM 343 O PHE A 43 2.361 28.823 34.393 1.00 12.52 O ATOM 344 CB PHE A 43 3.228 30.659 31.700 1.00 10.99 C ATOM 345 CG PHE A 43 3.993 30.709 30.401 1.00 9.80 C ATOM 346 CD1 PHE A 43 3.743 29.794 29.386 1.00 9.85 C ATOM 347 CD2 PHE A 43 5.032 31.615 30.233 1.00 11.37 C ATOM 348 CE1 PHE A 43 4.528 29.781 28.220 1.00 10.71 C ATOM 349 CE2 PHE A 43 5.816 31.612 29.075 1.00 10.61 C ATOM 350 CZ PHE A 43 5.569 30.697 28.067 1.00 10.48 C

PDB file format

Atom number

Residue type

Residue number X,Y,Z coordinates

Temperature factor

Atom identifier

ChainOccupancy Atom type

ATOM 340 N PHE A 43 3.853 28.346 32.161 1.00 10.57 N ATOM 341 CA PHE A 43 3.839 29.688 32.724 1.00 12.33 C ATOM 342 C PHE A 43 3.096 25.747 34.047 1.00 13.20 C ATOM 343 O PHE A 43 2.361 28.823 34.393 1.00 12.52 O ATOM 344 CB PHE A 43 3.228 30.659 31.700 1.00 10.99 C ATOM 345 CG PHE A 43 3.993 30.709 30.401 1.00 9.80 C ATOM 346 CD1 PHE A 43 4.743 29.794 29.386 1.00 9.85 C ATOM 347 CD2 PHE A 43 5.032 31.615 30.233 1.00 11.37 C ATOM 348 CE1 PHE A 43 4.528 32.781 28.220 1.00 10.71 C ATOM 349 CE2 PHE A 43 5.816 31.612 29.075 1.00 10.61 C

ATOM 350 CZ PHE A 43 5.569 30.697 28.067 1.00 10.48 C

12345678901234567890123456789012345678901234567890123456789012345678901234567890 1 2 3 4 5 6 7 8ATOM 340 N PHE A 43 3.853 28.346 32.161 1.00 10.57 N ATOM 341 CA PHE A 43 3.839 29.688 32.724 1.00 12.33 C ATOM 342 C PHE A 43 3.096 29.747 34.047 1.00 13.20 C ATOM 343 O PHE A 43 2.361 28.823 34.393 1.00 12.52 O ATOM 344 CB PHE A 43 3.228 30.659 31.700 1.00 10.99 C ATOM 345 CG PHE A 43 3.993 30.709 30.401 1.00 9.80 C ATOM 346 CD1 PHE A 43 3.743 29.794 29.386 1.00 9.85 C ATOM 347 CD2 PHE A 43 5.032 31.615 30.233 1.00 11.37 C ATOM 348 CE1 PHE A 43 4.528 29.781 28.220 1.00 10.71 C ATOM 349 CE2 PHE A 43 5.816 31.612 29.075 1.00 10.61 C ATOM 350 CZ PHE A 43 5.569 30.697 28.067 1.00 10.48 C

PDB files - problems

- PDB format uses fixed-width fields, so one entry is limited to 99,999 atom records and chain identifier is limited to single character (not even for structures of huge complexes - e.g. ribosome and viruses)12345678901234567890123456789012345678901234567890123456789012345678901234567890 1 2 3 4 5 6 7 8ATOM 340 N PHE A 43 3.853 28.346 32.161 1.00 10.57 N

- parsing of PDB files difficult - apart from ATOM records the file is almost unstructured (e.g. no rules to describe structure determination in REMARKS records)

mmCIF and XML formats deal with these issues

Trust PDB?

The database centres can’t refuse to accept any data! Even if curators of the PDB know the data contain serious errors.

So, PDB does contain a lot of errors - from sequence consistency errors (you’ll deal with them) to completely wrong folds.

And even the best data are still only the models that fit best experimental data.

Never trust the PDB!

Trp D 67 7GPB

Do you find this Trp normal?

Validation of structure files

- check statistics for bond lengths, angles, Ramachandran plots….

- do statistics look similar to those of other proteins?

WhatCheck, Procheck

- how well does the model fit experimental data?

EDS

Electron Density Server

PDBsum

PDBSum-Highlights

Text searches in structural databases

Options:

PDB - SearchLite, SearchFields

MSD - MSDlight, MSDpro (Java), MSDmine

OCA

Find all the structures deposited by Gerard Kleywegt with resolution better than 2Å and published in Journal of Molecular Biology

Search Fields

Summary

- three major repositories of structural data: RCSB, MSD and PDBJ

-all three are part of wwPDB

-structural data are deposited in PDB files - problems

- new formats - mmCIF, XML

- validation tools are necessary - WHATCheck, EDS

- new services are developed to analyze the whole database (MSD services)

- searches at various levels of depth/complexity - Searchlite, Search Fields

- added-value databases - OCA, PDBSum

Structural alignment

Why structural alignment ?

we have sequence alignment - Clustal…

KTHLCV

KSHA-V that gives us an idea about a correspondence of amino acids of two (or more ) proteins

That enables to infer information about function

And evolution of the Protein

If the sequences are similar enough !!!!

What is twilight zone ?

Sequence alignment unambiguously distinguishes only between protein pairs of similar structure and non-similar structures when the pairwise sequence identity is high.

High sequence identity roughly means over 40 %.

The signal gets blurred in the twilight zone of 20-35 % sequence identity.

More of the twilight zone

More than 90 % sequence pairs with the sequence identity lower than 25 % have different structures.

Significance of sequence alignments is length dependent.

The longer the sequence the lower identity is required to be called significant.Nevertheless, it converges to 25% with alignments longer than 80 amino acids.

‘The more similar than identical’ rule can reduce a number of false positives.

Using intermediate sequences for finding links between more distant families can also reduce the number of false positives.

How far can the sequence identity drop?

Average sequence identity of random alignments - 5.6 %

Average sequence identity of remote homologues 8.5 %

How does it work?

From http://www.biochem.unizh.ch/antibody/Introduction/Institutsseminar97/source/slide2.htm

Structural alignment because:

Structures are better conserved than sequences

structural alignment can imply a functional similarity that is not detectable from a sequence alignment .

Might help to improve sequence alignment when structures are available (phylogenetic studies, homology modeling).

Will improve sequence alignment methods (use of structural alignments’ substitution matrices, gap penalties).

Will improve sequence prediction methods

1FWR_A -------------------------MKNWKTSAESILTTGPVVPVIVVKKLEHAVPMAKA2YPI_A ARTFFVGGNFKLNGSKQSIKEIVERLNTASIPENVEVVICPPATYLDYSVSLVKKPQVTV ::. . . : :. * .. : . * ...

1FWR_A LVAGGVRVLEVTLRTECAVDAIRAIAKEVPEAIVGAGTVLNPQQLAEVTE-------AGA2YPI_A GAQNAYLKASGAFTGENSVDQIKDVGAKWVILGHSERRSYFHEDDKFIADKTKFALGQGV . .. . :: * :** *: :. : . :: ::: *.

1FWR_A QFAISPGLTEPLLKAATEGTIPLIPGISTVSELMLGMDYGLKEFQFFPAEANGGVKALQA2YPI_A GVILCIGETLEEKKAGKTLDVVERQLNAVLEEVKDWTNVVVAYEPVWAIGTGLAATPEDA . :. * * **.. : :.:.*: : : .:. :. .... :*

1FWR_A IAGPFSQVRFCPKGGISPANYRDYLALKSVLCIGGSWLVPADALEAGDYDRITKLAREAV2YPI_A QDIHASIRKFLASKLGDKAASELRILYGGSANGSNAVTFKDKADVDGFLVGGASLKPEFV * :* .. . * . : . ..: . .* * :.* * *

1FWR_A EGAKL--2YPI_A DIINSRN

Structural versus sequence alignment

Sequence 1 ------------ART---FFVGGNFKLNG-SKQSI-KEIVERLNTASI--PENVEVVICP .=ALI |=ID | |.... .. ..... . ....|... . | ... Sequence 2 MKNWKTSAESIL--TTGP--VVPVI--VVKKLEHAVP-MAKALVAG-GVR-----V-LEV

Sequence 1 ------PATYLDYSVSLV-KKPQVTVGAQ-N--AY-LKASGAFTGEN-S---VDQIKDVG .=ALI |=ID ...........| . ..|||. . . . . . .| Sequence 2 TLRTECAVDAIRAIAKEVP-E--AIVGAGTVLN-PQ----------QLAEVT--E---AG

Sequence 1 AKWVILGH--SERRSYFHEDDKFIADKTKFALGQGVGVILCIGETLEEKKAGKTLDVVER .=ALI |=ID |...|. . .....|.|.......|..|. ... Sequence 2 AQFAIS-PGL-------------TEPLLKAATEGTIPLIPGIS--------------TVS

Sequence 1 QLNAV-LEEVKDW-TNVVVAYEP--VW--AIGTGLAATPEDA--QDI--HASI-RKFLA- .=ALI |=ID .|... . .. . .....| . . . . .. . . Sequence 2 ELMLGMD--YG-LK---EFQFFPAE-ANG-------G----VKA--LQA--IAG-P--FS

Sequence 1 SKLGDKAA-SELRILYGGSANGSN-AVTF---KDK-ADVDGFLVGGA-SLK--------- .=ALI |=ID . |....|... .. . . . ..|..... .. .. Sequence 2 -------QV---RFCPKGGIS-PANY--RDYL--ALKSVLCIGG-SWL-VPADALEAGDY

Sequence 1 --P--EFV--DIIN--SR-N .=ALI |=ID . . . . . .. Sequence 2 DRITKL-AREA--VEGAKL-

1 2 3 4 5 6 7 8 9 10 11 12 13 14 PHE ASP ILE CYS ARG LEU PRO GLY SER ALA GLU ALA VAL CYS PHE ASN VAL CYS ARG THR PRO --- --- --- GLU ALA ILE CYS PHE ASN VAL CYS ARG --- --- --- THR PRO GLU ALA ILE CYS

Sequence versus structural alignment

Is it difficult to make structural alignment?

Structural alignment is NP-hard (nondeterministic polynomial time) problem.

In other words, it is not tractable properly.

Even, if it would, the result would be correct from technical point of view not necessary from biological point of view.

Yes, it is.

General solution

Use a heuristic approach:

1. Represent the proteins A and B in some coordinate independent space

2. Compare A and B

3. Optimize the alignment between A and B (e.g. minimize R.M.S.d.)

4. Measure the statistical significance of the alignment against some random set of structure comparisons

“..in some coordinate independent space…”

Make the problem easier by:

- comparing only distance matrices of atoms

-comparing secondary

structure element (SSE)

- comparing cartoons

- comparing vectors of SSE

- combination of mentioned methods

- ….

None of the methods guarantee the finding of the closest structure and two methods can disagree at all amino acid positions.

Nevertheless they can still provide a valuable insight into the history of the protein and give hints concerning the function.

Server Location MethodCE http://cl.sdsc.edu Extension of optimal path1

DALI http://www2.ebi.ac.uk/dali Distance-matrix alignment2

DEJAVU http://portray.bmc.uu.se/cgi-bin/dennis/dejavu.pl SSE alignment with Catom optimisation3

LOCK http://gene.stanford.edu/LOCK/ Absolute orientation of corresponding points4

MATRAS http://bongo.lab.nig.ac.jp/~takawaba/Matras.html

Markov transition model of evolution5

PRIDE http://hydra.icgeb.trieste.it/pride/ CCatom distances6

SSM http://www.ebi.ac.uk/msd-srv/ssm/ssmstart.html Graph matching algorithm

TOP http://bioinfo1.mbfys.lu.se/TOP SSE alignment7

TOPS http:// tops.ebi.ac.uk/tops/compare1. html TOPS-diagram alignment8

TOPSCAN http://www.rubic.rdg.ac.uk/~andrew/bioinf.org/topscan

Secondary topology-string alignment9

VAST http://www.ncbi.nlm.nih.gov/Structure/VAST/vastsearch.html

Vector alignment10

Methods for fold comparison

Protein structure classification

If you want to know which structures are similar to a known structure, these systems might help:

A) Manual - SCOP

B) Semi-automatic - CATH

C) Automatic - FSSP

CATHCATHhttp://www.biochem.ucl.ac.uk/bsm/cath

CATH Topology or fold group level From C. Orengo talk at EMBO course, Cambridge 2004

TIM barrel enzymes – 18 different homologous families

>60 different E.C. numbers

EC Wheel of TIM barrelsStructure of TIM barrel:Triose phosphate isomerase

From J. Thornton talk at EMBO course, Cambridge 2004

Rossmann Fold

Jelly Roll

Alpha/Beta Plaits

Arc repressor-like

OB Fold

CATHCATH

Rossmann

Alpha-beta plait TIM barrelJelly Roll

Immunoglobulin

OB fold

SH3-like

Up-down

Arc repressor-like

nearly one third of the

superfamilies belong to <10 fold

groups

From C. Orengo talk at EMBO course, Cambridge 2004

TargetDB

contains 84063 sequences annotated like:-hypothetical protein Af0491 from A. fulgidus

- putative serine hydrolase from S.cerevisiae

-predicted glutamine amidotransferase from P. aeruginosa (January 2005)

QuickTime™ and a TIFF (Uncompressed) decompressor are needed to see this picture.

http://targetdb.pdb.org/

PDB contains about 500 structures with a similar degree of confidence in functional assignment

Hypothetical Protein Mth938 (PDB ID:1ihn)

Function from structure

Fold and structural motifs

SSM fold search

Surface clefts

Residueconservation

DNA-bindingHTH motifs

Nest analysis

Sequence motifs(PROSITE, BLOCKS,SMART, Pfam, etc)

Sequence scans

Sequence searchvs PDB

Sequence searchvs Uniprot

Superfamily HMMlibrary

Gene neighbours

n-residue templates

Enzyme active sites

Ligand binding sites

DNA binding sites

Reverse templates

Summary

Structural alignment can help with protein annotations even when the sequence similarity is not significant.

Sequence identity of two proteins with similar structures can be lower than 10 % - number of folds is limited.

Recent progress in protein structure determination increases the usefulness of structural alignment.

Structural alignment is difficult problem that is solved by heuristic methods.

These methods simplify the problem and sacrifice the optimum result for the speed.

Summary II

Different methods can provide completely different alignments.

In our results, CE, Dali,Matras and Vast were the best servers for finding structural relatives.

A few structural classification systems have been developed (CATH, FSSP, SCOP), they provide hierarchical classification of protein structures and enable to infer functional and evolutionary relationships between proteins.

Folds are not distributed equally. Ten most frequent folds represent almost one third of all structures.