1 szabolcs csepregi*, szilárd dóránt, nóra máté, miklós vargyas, péter kovács, györgy...

1

Szabolcs Csepregi*, Szilárd Dóránt, Nóra Máté, Miklós Vargyas, Péter Kovács, György Pirok, Ferenc Csizmadia

First presented at Applications of Cheminformatics and Chemical Modelling to Drug Discovery · 8-19 Nov 2004 Updated. April, 2005

Structural Search Using ChemAxon Tools

Structural Search Using ChemAxon Tools — April 2005 2

Contents

Structural search in cheminformatics

The JChem suite of tools

Structural search in JChem• Interfaces

• Database solutions: JChemBase, Cartridge

• Standardization

• Search features

MCS/MCES and Library MCS

R-group decomposition

The Chemical Terms language

Future plans

All examples generated by ChemAxon’s Marvin


Structural search in cheminformatics

A few examples to highlight the diversity of applications. :

•Compound registration – duplicate checking

•Database search e.g. imidazole derivatives

•Pharmacophoric group identification (JChem Screen, JKlustor)

•Functional group identification

•Cleavage bond identification (JChem Fragmenter)

•Virtual reaction processing (JChem Reactor)

•Standardization (canonicalization of structures, JChem Standardizer)

•Toxical fragment identification (superstructure search)


Search types in JChem

• ABAS(Atom By Atom Search) or structural search:– Exact– Substructure– Superstructure– MC(E)S – maximum common (edge) substructure– R-group decomposition (identify ligands of a given scaffold)

• Similarity search:– Different Descriptors– Different Metrics


ABAS search interfaces

• JSP(Java Server Pages): web GUI for database– Similarity & structural search– Substructure highlighting– Additional constraints– Insert, modify, delete

• Command line utility: jcsearch: for files and DB

• Java API– isMatching() – Only to check matching– findFirst(), findNext() Enumerate all– findAll() possible matchings

• Cartridge: access all functionality from SQL

• Chemical Terms


ABAS options

• General options:• Order sensitive hits e.g.• Pre-assignment of query and target atoms• Consider stereo or not, absolute stereo (ignore chiral flag)• Timeout limit• Exact charge/radical/isotope/query features/bond/stereo matching• Double bond stereo: no check/marked/all double bonds• Chemical Terms filter expression• etc

• Database search:• Maximum search time/number of hits• Additional SQL SELECT expression for prefiltering• Output table• Reverse hits mode


Structural search in database

• Search: two stage method:– Rapid pre-screening based on chemical hashed fingerprints– ABAS (isMatching)

• Duplicate check at compound registration:– Hash code: primary filter– ABAS (isMatching)

• Standardization• Caching of structures and fingerprints allow top performance


Import with JChem Base Manager


JChem Base molecular file formats and integration

Import formats:• SMILES• MDL molfile (v2000 and v3000)• MDL SDF• RXN• RDF• MRV

Database engines:• Oracle• MySQL• MS SQL Server• PostgreSQL• MS Access• DB2• etc.

• CML• PDB• Sybyl molfile• XYZ• Gaussian cube• Image formats for export(JPG, PNG, SVG)

OS: any operating systems running java

• Windows• Linux• Mac OS X• Solaris• etc.


JChem Base performance (1)

Compound registration:

Substructure search in a table of 3 million compounds:

Server parameters: Windows XP; 1 CPU: Intel P4 3.0GHz; 2GB RAM; Oracle 9i

Number of compounds

Elapsed time

Duplicates not checked Duplicates checked

10,000 32s 45s

100,000 4min 11s 6min 20s

200,000 8min 17s 12min 26s

10.749740

1.20

0.9936

0.112

Search time (s)Number of hitsQuery


JChem Base performance (2)

Similarity search:Tanimoto >0.8

Server parameters: Windows XP; 1 CPU: Intel P4 3.0GHz; 2GB RAM; Oracle 9i

1.3336

1.3156

1.524

Search time (s)Number of hitsQuery


JChem Cartridge for OracleJChem Cartridge for Oracle


JChem Cartridge for Oracle

Oracle is extended to support chemical database operations using the JChem Cartridge for Oracle

Examples:

Substructure search displaying ID, SMILES codes, and molweight:

SELECT cd_id, cd_smiles, cd_molweight FROM my_structures

WHERE jc_contains(cd_smiles, 'CC(=O)Oc1ccccc1C(O)=O') = 1;

Similarity search filtered with predicted pKa values, which displays predicted logP and logD values:

SELECT cd_id, jc_logP(cd_smiles), jc_logD(cd_smiles, 7.4) FROM my_structures

WHERE jc_tanimoto(cd_smiles, 'CC(=O)Oc1ccccc1C(O)=O') >= 0.8

AND jc_pKa(cd_smiles, 'acidic', 1) < 4;




Chemical Terms examples:• Number of compounds in table “nci_10m” containing benzene and conforming the Lipinski rule of 5:

SELECT count(*) FROM nci_10m WHERE jc_compare(structure,'c1ccccc1','sep=! t:s!ctFilter:(mass() <= 500) && (logP() <= 5) && (donorCount() <= 5) && (acceptorCount() <= 10)') = 1

• Compounds in table “nci_10m” containing 3-bromoindole and restricting TPSA, molecular weight, rotatable and aromatic ring counts:

SELECT cd_structure FROM nci_10m WHERE jc_compare(structure,'Brc1cnc2ccccc12','sep=! t:s!ctFilter:(PSA() <= 200) && (rotatableBondCount() <= 10) && (mass() <= 500) && (aromaticRingCount() <= 4) ') = 1

• New interface to ChemAxon API features from SQL — accessible from non-java programs as well.

• Enhanced performance of certain SQL queries.



Query features 1. Atomic features

• Query atom types: any, hetero, list, not list

• Pseudo atoms e.g. “Resin”

• Explicit lone pairs (matches to implied lone pairs as well.

• Charge, isotope, radical

• Query properties:

Symbol Description

H<n> Total hydrogen count

a Aromatic

A Aliphatic

R<n> Ring count in SSSR

r<n> Ring size in SSSR

v<n> valence

X<n> Connectivity


Query features 2. Atomic SMARTS features

• SMARTS atoms:

• Additional query properties:

• Example:

Carbonyl C, but not amide

Symbol Description

D<n> Degree

h<n> Implicit H count

& ; , ! Logical operators

$(<smarts>) Recursive smarts

+0, -0 Zero charge


Query features 3. Bond features & components

• Query bond types: Any, single or double, single or aromatic, double or aromatic

• Bond topology: chain/ring

• Smarts bonds

• Component level grouping

Symbol Description

- = # Single, double, triple

: aromatic

& , ; ! Logical operators

@ Ring bond

/ \ /? \? Directional bond (cis/trans)

Symbol Description

(C.C) Same component

(C).(C) Different component

C.C No component restrictions


Stereo searching 1. Double bonds

• Levels of check:– All– Only marked double bonds

(MDL: stereo care flag)

– None

Not cis

Not trans

Cis or trans

(unknown)

Trans

Cis

MeaningDepiction


Stereo searching 2. Tetrahedral chirality

• Stereo bond types:

• Relative stereo configuration• Chiral flag model• Enhanced stereo representation: AND<n>, OR<n>, ABS groups

Up or downDownUp


Reaction search

• Reactants, agents, products

• Transformation recognition (mapping)

• Stereospecific reactions (inversion, retention)

• Reactant grouping


R-group search

• Scaffold, R-group definitions

• Monovalent, divalent R-groups

• R-logic

•Occurrence

•If-then

•RestH


Hydrogens

• H representations:– Explicit– Implicit– Query H count (total or implicit)

• Example:

Considered in ABAS

Explicit H Implicit H Query H count

Query

Target

Target

Query


Standardization

• Explicit hydrogens removal

• Aromatic bonds

• Mesomers

• Tautomers

• Counterions

• Stereo representation


Standardization - Aromaticity

• RepresentationsKekulé Aromatic

• Example: The two Kekulé representations below don’t match

• Two options available: ChemAxon & Daylight aromatization


Standardization Example

afterafterbeforebefore


Similarity search

• Descriptors:– Chemical hashed fingerprint

– 2D (topological) pharmacophore fingerprint

– BCUT

– Structural keys

– Hypothesis fingerprints: minimum, average

• Dissimilarity Metrics:

– Tanimoto: standard, scaled, asymmetric

– Euclidean: standard, normalized, weighted, asymmetric

– Optimized for a set of actives


MC(E)S 1. Pairs of molecules

•The largest connected commonsubgraph

•Application: reaction automapping in Marvin


MCS 2. Library MCS

The LibMCS program rapidly creates a hierarchy of MCS-es on a library.

Applications:• Identification of the

most frequently occurring MCS.

• Focused set analysis• Clustering based on

common substructures


Hierarchy calculation performance

Library Library size

Time(s) Clusters Top level clusters

No. of levels

NCI (small molecules, random, diverse sets)

500 6.8 279 14 5

1,000 13.4 440 16 6

5,000 141 851 42 5

d2 inhibitors (medium sized molecules, low diversity)

500 10.8 243 3 7

1,000 25.6 495 9 6

Thrombin inhibitors (medium sized molecules, medium diversity)

1,000 67 528 8 6

3,000 242 1186 12 6


R-group decomposition

JChem is able to identify the ligands of a given scaffold at specified substitution positions:

Query(scaffold) Result

Library R-groupdecomposition


Applications of Chemical Terms

CT

virtual synthesisreaction and synthesis rules

pharmacophore analysispharmacophore definitions

drug designgoal functions

structural searchadvanced query expressions

e.g. in the Cartridge


Chemical Terms

searching match("olefine.mol") && !match("c1ccncc1") && (atomCount(16) == 0) || (mass() < 300);

goal functions inhibitor = inhibitor.mol;

(similarity(inhibitor, pharmacophore_tanimoto) > 0.8) && (similarity(inhibitor, chemical_tanimoto) < 0.5);

filtering (mass() <= 500) &&

(logP() <= 5) &&

(donorCount() <= 5) &&

(acceptorCount() <= 10);

• structure matching functions (describing functional groups, reaction sites, similarity…)

• property calculations (partial charge distribution, pKa, logP, electrophility…, etc)

• arithmetic and logic-operators

Elements of the language

Chemical Terms examples


Chemical Terms

Some available functions

• Structural search (match, matchcount)• Partial charge distribution • pKa, Log P, Log D, major microspecies• Polarizability• Topological Polar Surface Area• Number of rotatable bonds, rings, aromatic rings, etc.• Number of HB donors/acceptors• Exact mass • Arithmetic and logic operators • Extensible: your own Java plugins can be easily added.

• Etc.


Future plans

• More query features (e.g link nodes, ring bond count, unsaturated atom)

• Flexible search options: tautomeric search, ignore bond types, salts, etc.

• Search targets having R-groups (Markush structures)

• etc.


Summary

Structural search provides a useful set of tools for chemists and cheminformaticians.

ChemAxon JChem suite contains a broad range of chemical search facilities and the presented benchmark results illustrate the high performance of JChem search.

The new Chemical Terms language is a beneficial complement to structural searches allowing data mining made easy.


Links

• Home page– www.chemaxon.com

• Forum– www.chemaxon.com/forum

• Animated demos and tutorials– www.chemaxon.com/demos

• Presentations and posters– www.chemaxon.com/conf


Máramaros köz 3/a Budapest, 1037Hungary

[email protected]

www.chemaxon.com

Thank you for your attention

1 szabolcs csepregi*, szilárd dóránt, nóra máté, miklós vargyas, péter kovács, györgy...

Documents