of the Purdue Ontology for Pharmaceutical

sounderslipInternet and Web Development

Oct 22, 2013 (3 years and 7 months ago)

69 views

Excipient Interaction Prediction: Application
of the Purdue Ontology for Pharmaceutical
Engineering (POPE)


L. Hailemariam, P. Suresh, P. Akkisetty, G. Joglekar,

S.
-
H. Hsu, A. Jain, K. R. Morris, G. V. Reklaitis, P. Basu,

V. Venkatasubramanian

School of Chemical Engineering, Purdue University

ESCAPE 18, Lyon, France

June, 2008

2

Drug Product Stability


A drug product includes the active pharmaceutical ingredient (API)
and ‘inactive’ excipients


Primary drug degradation reactions are hydrolysis, oxidation and
photolysis


Affected by Temperature, pH, etc. (environmental conditions)


Reaction prediction systems


Computational Chemistry (Mechanistic Computation)


SPARTAN (Wavefunction): transition state geometry


Expert Systems: MetabolExpert


Database Access:


ROBIA (Reaction Outcomes By Informatics Analysis): rules


DELPHI (Degradant Expert Leading to PHarmaceutical Insight)


Little use of reaction environment information in reaction prediction


Challenge of integrating different forms of information

3

Information Structure


Ontology is the explicit description of domain concepts and relationships
between concepts


Water

Boiling point

100
0
C

has

Material

Property

Value

has

has


Ontology captures consensual knowledge, data representation and logic


Ontology Language: Web Ontology Language (OWL)


Ontology =

a

of

4

Purdue Ontology for Pharmaceutical Engineering


Knowledge Representation


Guideline knowledge: actions, decisions
1


Mathematical knowledge: math models (equations, assumptions, etc.)
2


Information Model


Phase description (phase, composition)


Molecular structure description (fragments, connection)


Reactions (model, participants)


Material properties (density, flow, heat capacity)


Experiments (procedure, settings)


Unit operations (streams, interaction)


Equipment and Value




Superset = Purdue Ontology for Pharmaceutical Engineering (POPE)

1
Poster session

2
Session 5.3, 11:20, June 4 (Wed.)

5

POPE Overview

(A)
Information Modeling (POPE
-
Im)

(B)
Knowledge Modeling (POPE
-
Km)

(C)

Mathematical Modeling (POPE
-
Mm)

6

Applications


Decision Support for Product Development
1


Decision steps modeled explicitly


Ontology instances accessed for property values




Mathematical Modeling of Unit Operations
2


Input/output, assumptions, solution method described
explicitly


Ontology instances accessed for property values



Reaction Prediction


Instance comparison, reasoning, database link

1
Poster session

2
Session 5.3, 11:20, June 4 (Wed.)

7

Proposed Reaction Prediction System



An ontology
-
based approach proposed to assist reaction prediction


Ontology supports both data access and reasoning


Which reaction may occur and why


Link to relational database developed



Source Reaction databases: MerckIndex, SigmaAldrich, Metasynthesis



CDK tools used to identify presence of common fragments

8

Reaction Prediction Application

Molecular Structure Information

[SMILES string]

e.g. NC1ONC1=O

Environmental Information

[text]

e.g. T=25 C

Deterministic Search

[Rule engine]

e.g.
Substance (Cycloserine)

^has_fragments(Cycloserine, ?a)

^participates_in_rxns(?a,?b)

^
has_temperature
(?b,?c)

Score
-
based Search

[Molecular and environmental similarity]

e.g.
Tanimoto Score

DEMO

Result



Test case: 10 blockbuster drugs in accelerated testing


50% of predicted reactions found in literature


73% of known reactions predicted

Molecular Structure Ontology

Reaction Ontology

9

Reaction Score Table for Cycloserine

10

Result


Predicted reactions compared with open literature


50% of predicted reactions found in lit.; 73% of reported reactions predicted



* O = predicted, not reported in literature;


= predicted and reported
;
x = reported in literature

11

Summary


Presence of undesired reactions a major problem for pharmaceutical
products



An ontology
-
based approach developed to assist information
integration at multiple levels and reasoning for reaction prediction



The Purdue Ontology for Pharmaceutical Engineering (POPE)
developed to describe materials, chemical structures, reactions,
material properties and experiments



The system predicted most of the reported reactions for the test set




Future work includes incorporation of more chemistry descriptors and
evaluation of multiple similarity measures

12


Appendix

13

Score
-
based reaction prediction


Considerable granularity in reaction prediction.


Secondary interactions (van
-
der
-
Waal ) affect reactivity




S
uggest reactions, ranked by likelihood/similarity to new environment



Rxn similarity mostly based on molecule similarity, Tanimoto Measure used

a, b= # of fragments of molecule 1&2;

c= number of common fragments


Structural similarity (Fragment*Ring): common fragments, ring size coincidence


Environmental similarity: Phase, Temperature, pH, RH, PSD range overlap


Structural Match Score = Fragment Match Score*Ring Match Score


Total Match Score = Structure Match Score*Environment Match Score

14

10/22/2013

Purdue Ontology for Material Entities (POME)


Previous Work
: Model.LA, ISO 10303, OntoCAPE


Example
:
Substance
: H
2
O
Phase system
: ice, steam


Application
: reaction prediction, formulation decision support system

15

Purdue Ontology for Molecular Structures (POMS)


Previous Work
: ChEBI, Hsu et al (2006)


Constructed by looking at patterns in drug degradation reactions


32 fragments, rings: checkmol list (156) PubChem (880)


Example: Cycloserine (Substance) has carbonyl, ether, amine Fragments


Application
: reaction prediction

16

10/22/2013

Purdue Ontology for Reaction Expression (PORE)



Previous Work
: Hsu et al (LIPS)


Example
: hydrolysis, oxidation; polymorphism, melting, condensation


Application
: reaction prediction, unit operation modeling

17

10/22/2013

Purdue Ontology for Material Properties (POMP)


Previous Work
: OntoCAPE, ISO10303, Model.LA


Little treatment of solid properties, lack of integration with experiments


Example
: angle of repose, powder density, heat capacity


Application
: formulation decision support system, unit operation modeling

18

Property Hierarchy



Phase System Properties


Interphase Properties


Solid Properties


Particle Properties


Crystalline Properties


Powder Properties


Mechanical Properties


Micromeritics


Thermodynamic Properties


Energetics


Physical Chemistry



Substance Properties


Chemical Constants


Molecular Properties


19

10/22/2013

Purdue Ontology for Description of Experiments
(PODE)


Previous Work
: EMB, GAML


Lack of integration with properties


Example
: Hausner Ratio measurement, HPLC


Application
: experiment analysis

20

Unit Operation Ontology


Previous Work
: Model.LA, ISO 1303, OntoCAPE


Lack of integration with experiments


Example
: boiling, drying


Application
: unit operation modeling

21

Equipment Ontology


Previous Work:

CLiP, Sunagawa et al (2003), Lohse et al (2006)


Example
:
Reactor

has
inlet port
,
outlet port
; has volume 1 m
3


Application
: experiment analysis, unit operation modeling

22

Value Ontology


Previous Work
: EngMath, ISO 10303


Developed primarily for math modeling


Example
: Angle of Repose has
Range [51,54]


Application
: formulation decision support, unit operation modeling,
reaction prediction, experiment analysis