Daylight/Oracle Chemistry Cartridge

Daylight/Oracle Chemistry Cartridge
Using the Chemistry Cartridge

Alberto Gobbi

Novartis Crop Protection

Why do we need it?

Daylight is a great Chemical Information System!

But not a perfect database system!

For data modeling.

For application development.

For database management.

Data Modeling: How do we Create Data Models?

$D<"$SMI">
_V<Structure:SMILES>
_B<SMILES>
_N<USMILES MAKEGRAPH>
_P<*>
_S<SMILES notation, primary identifier in the Daylight system>
_D<The uniquified SMILES notation is the primary key to all data in the Daylight Chemical Information System. See the Daylight Software Manual for a comple
te description.>
_M<System, Structure, Lookup, Common>
TS<199707081541.19>
|
$D<"$ISW">
_V<"indirect/keywords;keywords">
_B<"$ISW/id;$ISW/data">
_S<Indirect reference for keyword data, used in Spresi datatypes JA and PAT>
_P<>
_C<Spresi datatype>
_M<Indirect>
_O<mjollnir@daylight.com>
TS<199707081541.19>
|
$D<AMW>
_V<Ave molecular weight>
_B<Ave MW>
_N<NUMERIC>
_P<>
_S<Average molecular weight using isotope distribution in nature>
_D<Average molecular weight using isotope distribution in nature>
_M<System, Medchem, Measured>
TS<199707081541.19>|

hview nci3d

comment This is the Hview for the NCI-3D 94.1 database with MODEL over MOL.
comment 11/16/94

tree mol
device maccsdb
database nci3d:
tname mol
password
rename mol>(molclass) to <hide>
rename mol>(molskeys) to <hide>
rename mol>(molname) to <hide>
tinfo key molregno

tree model basedon mol
tname model
password
rename model>(molregno) to (2d_regno)
rename model>(modelname) to <hide>
tinfo fieldtype source model.source flatten text
tinfo key modelregno

link model model>(molregno) over mol (molregno)

Data Modeling: How is your Data Model?

Data Modeling: A Relational Model

Why? For Data Modeling!

Get all the nearest neighbors of

SELECT s.Structure FROM Structures s, Neighbors n WHERE
s.id = n.Neighbor AND n.Compound = 3578;

Get all salts of

SELECT s.Structure FROM Structures s, Salts sa WHERE
s.id = sa.salt AND sa.BaseCompound = 1001;

All salts with same BaseCompound.

Different depiction (display formats) for different users?

Link biological, toxicological, physical or Spectroscopic data

Update Neighbors table by triggers.

llll

Application Development:
A 100% Daylight Database Example

Easy Fullstructrue Search.

No enforcement of relationships in the database.

No consistency checks in the database.

Complicated navigation through the data hierarchy.

No link to biological data.

No transactions.

No data aware controls (Java Beans, ActiveX).

Application Development:
An Oracle 7/Daylight Toolkit Example.

No substructure search.

Complicated SMILES canonization.

Complicated structure normalization.

No tautomer search.

No hash lookup.

No similarity search.

Application Development:
An Oracle 8I Example.

Includes Daylight SSS.

Includes similarity search.

Implementation effort: 5H

Easy SMILES canonization.

Callouts for structure normalization.

Why? For Application Development

Enforce data consistency.

Enforce relationships.

Allow for transactions.

Simple query and update of any fields via SQL.

Easy link to any other data in house.

Standard Client Server interfaces (JDBC, ODBC).

Data aware controls (Java Beans & ActiveX).

Fullstructure, substructure, tautomer and similarity search.

Why? For Database Management

Current structural input of 500 000 compounds per year.

TPO Third Party database 2 000 000 records.

Screening Database 500 000++ entries.

Future input 1 000 000++ compounds per year.

Cartridge advantages:

Maintenance by professionals.

Partitioned backup.

On-line backup.

Parallelization.

Scaleable?

First Experience: Not Perfect but Alive!

Searches work as advertised:

SSS on 475 000 structures in 15 to 120 sec (no caching).

NNS or Similarity Search on 475 000 structures in 5-10 sec.

75 – 100 MB per session and 475 000 compounds.

NN Search does not work with joins.

Joins currently require a complete scan on the chemistry table.

No Tversky similarity.

Indexing is really slow:

54 H for 470 000 structures.

80 sec to add one compound.

>6 Days for 700 000 structures.

>800 MB for 700 000 structures.

Do we need more?

Requirements:

Oracle optimizer to allow for fast joins.

Access to some toolkit functions like dt_cansmi, dt_match dt_fp_teverski.

Functions to calculate molecular formula and weight.

Conversion function from and to SMILES and MOL-Files in the database, (Bernds 2D coordinate generation).

Check for corporate structural rules in the database (API).

Bug fix in SMILES canonization.

Nice to have:

Direct MERLIN load from database.

Update of MERLIN pool via database triggers.

DayBlob is Alive!

It is not perfect but a large step ahead!

Daylight

Norah MacCuish

Dave Weiniger

Oracle

Sam Defazio

Johnny Petersen

Cathleen Trezza

M. Visscher

Novartis

Paul Gatewood

Matthias Ochs

Dieter Poppinger

Bernd Rohde

CP_IT Department