In silico Prediction of Forced Degradation - Lhasa Limited an Expert... · pH profile from preceding slide . Knowledge Sources . General Pharmacological and Pharmaceutical Journals

Title

Martin A. Ott Lhasa Limited www.lhasalimited.org

In silico Prediction of Forced Degradation

Building an Expert Computer System

to Predict Degradation Pathways

Forced Degradation Studies, 27-28 January 2010 – Renaissance Hotel, Brussels

• Introduction

• Degradation prediction software

• Chemical knowledge base of transformations

• Scope and limitations

• New developments

• Information sharing and confidentiality

• Conclusion

Contents

Lhasa Limited is a not-for profit organisation that promotes knowledge and data sharing in chemistry and the life sciences

What is Lhasa Limited?

• Drugs (formulated or not) are exposed to harsh conditions to study their degradation behaviour

• Structural identification of degradation products

• Elucidation of degradation pathways

• Educated guesses on degradation are welcome

• Plenty of information available but very dispersed

Need for a predictive (expert) system

Forced Degradation

Drug Degradation Database (D3) * No prediction * Limited size http://d3.cambridgesoft.com/

CAMEO (reaction prediction) * No longer available * Not adaptable to specific needs Pure Appl. Chem. 62, 1921-1932 (1990) J. Org. Chem. 60, 490-498 (1995)

Delphi (degradation prediction) * In-house Pfizer project Mol. Pharm. 4, 539-549 (2007); DOI 10.1021/mp060103+

Degradation Software

A computer program that predicts chemical reactions needs to:

Predicting Reactions

• Understand chemical structures – chemistry engine

• Know chemistry – knowledge base

• Assess reaction likelihoods under different conditions

• Assess competition between reactions

• Hydrolytic

• Oxidative

• Photochemical

H2OO

R RR

OH

OH

R

RN

R

R

RN

+

R

ORROOH

R O

OR

R OH

O

OHR

H2O+

R

RO

R

RO

R

R

R

R 1O2+

RBr

RHhν R

RR

Rhν

Degradation Chemistry

• In Zeneth’s knowledge base, chemical reactions are represented through patterns, e.g.:

• The pattern defines both the transformation and the scope

N

OO

R1

R2

R3

N

OO

R1 R2

R3

O21

NR1

R2

R3

*

R1-R3 = aliphatic carbon (not multiply bonded to a heteroatom) or aromatic carbon or hydrogenThe bond marked * must be fused to another aromatic ring

Chemical Patterns

• Heat (temperature) #

• Acid & base catalysis (pH) #

• Hydrolysis (H2O)

• Molecular oxygen (O2)

• Peroxides

• Radical initiator

• Metal (Fe[III] or Cu[II])

• Photochemical (light)

Reaction Conditions

# = numerical; others indicate presence/absence

Reasoning

Seven likelihood levels are used:

Absolute reasoning: Determine the likelihood of transformations

Relative reasoning: Assess competing transformations

• (Certain) • Very likely • Likely • Equivocal • Unlikely • Very unlikely • (Impossible)

Conditions / Reasoning

Oxidative and photochemical reactions:

• Presence of a specific oxidant (or light) is a prerequisite for setting the likelihood level

• Any combination of conditions can be used

• Examples:

“S-Oxidation of thioethers is very likely when

either O2 or peroxides are present”

“Oxidation at benzylic positions is likely when

O2 and a radical initiator are both present”

Conditions / Reasoning

Hydrolysis reactions:

• Water is a prerequisite

• Likelihood of many reactions is dependent on pH

• Reactions that are both acid- and base-catalysed display a minimum in the pH-dependency

• Example of a pH profile: pH < 6 VERY LIKELY pH = 6-8 LIKELY pH = 8-10 EQUIVOCAL pH = 10-12 LIKELY pH > 12 VERY LIKELY

Conditions / Reasoning

Various pH profiles:

pH profile from preceding slide

Knowledge Sources

General Pharmacological and Pharmaceutical Journals Eur. J. Pharm. Biopharm. Int. J. Pharm. J. Pharm. Biomed. Anal. J. Pharm. Sci. J. Pharm. Pharmacol. Pharm. Res.

Editors: Dinos Santafianos (Pfizer) Steve Baertschi, Pat Jansen (Eli Lilly)

Knowledge Base Editor

Name Description Comments

Knowledge Base Editor

Transformation Attributes

R-group definition

Hydrolyses

Oxidations

Condensations/additions

Eliminations

Isomerisations/rearrangements

Photochemical reactions

Total

Knowledge Base Status

30

33

16

9

12

9

109

Sample Degradation

Hydrolysis

OO

OO

O ON

NN

O

N

OO

O

O

O

Oxidation

Hydrolysis

Oxidation

Degradation sites of rifampicin

Sample Degradation

Hydrolysis

OO

OO

O ON

NN

O

N

OO

O

O

O

Oxidation

Hydrolysis

Oxidation

Zeneth predictions (pH 7, water, oxygen, peroxide, one step):

Likely

Likely

Very likely Likely One more reaction

at the equivocal level

Degradation prediction of nordazepam

Sample Degradation

Degradation prediction of nordazepam

Sample Degradation

Scope and Limitations

Prediction of degradants as a result of:

• Shelf life time or stability studies

• Accelerated degradation studies (e.g. 2 months at 75% humidity)

• Forced degradation studies (e.g. O2/AIBN, 1 hour at pH 1)

Quantities, reaction rates

Likelihood of degradant formation

No

No

Yes

No

Yes

New Developments

New developments in 2009:

• Prediction of intermolecular (bimolecular) reactions

• Handling of chemical structures with radicals

• Support for more structure editors

• Continuous growth of the knowledge base (50 109)

Bimolecular Reactions

• One query compound is considered to be the “primary query compound” = Q (typically the API)

• Additional compounds entered are considered to be the “secondary query compounds” = A, B, … (typically excipients, counterions etc. but can also be another API)

• Intermolecular reactions are predicted between Q and A, Q and B, etc. but not between A and B, etc.

• Dimerisations (and polymerisations) are predicted when A is the same compound as Q.

Bimolecular Reactions

The knowledge base currently contains four intermolecular transformations

O

OH

O

N

OH

O

Ph

NH2Ph

O

NH

O

Ph

O

OH

ONH

Ph

O

OH

O

OH

O

O

OOH−

[ O ]

+

Q

Q

A

Reactants

Currently three classes of “secondary query compounds” have been identified:

• Excipients e.g. fructose, triacetin, aspartame

• Counterions e.g. succinate, citrate, maleate

• Contaminants (impurities from excipients including degradants) e.g. formaldehyde, glyoxal

Reactions Involving Radicals

Full support for radical structures has been added.

Radical compounds mainly occur as intermediates: Radicals in query compounds and product structures are supported as well.

C

OO

O2R RH

O

OOH

OH

RH

Alternative Structure Editors

In addition to ISIS/Draw, two more structure editors are now supported: Symyx Draw and ChemDraw.

• More chemistry − 160 transformations by the end of 2010

• Fine-tune likelihoods − through feedback from users

• Experimental data for examples

• More literature references

Work in Progress

Use of physicochemical properties to enhance predictions:

• pKa to assess protonation state

and deprotonation reactions

• bond dissociation energies to assess H abstraction reactions

• HOMO and LUMO energies

Plug-in calculators will be used that interface with the knowledge base

Work for the Future

Data Sharing

• A collaborative group has been set up

• Currently four members: Amgen Eli Lilly GlaxoSmithKline Johnson & Johnson

• Members co-direct development

• Handling of confidential data Transforming confidential data into non-confidential knowledge

Data Sharing

• Contributions from members Compound/degradation profiles New transformations / literature references

• Partial structures are often sufficient to describe the chemistry Unless the transformation is specific to a confidential scaffold

• Confidentiality status is covered by project agreement

• Data can be shared at different levels Fully public Anonymous

Benefits of Participation

Impact:

• Improve assessment of drug candidate stability through faster identification of degradation pathways

• Minimise studies for related compounds • Education and training of individuals • Potential to build and maintain institutional

knowledge

Benefits of Participation

• A strong team contributing chemical knowledge • Careful testing against actual pharmaceutical

models • Suggestions for functionality to meet industry

needs

• Improving the system over time • Maintaining the software over time and across

platforms

Development - going beyond basic functionality requires: Sustainability - the consortium model provides stability and a mechanism for:

Conclusion

• Development on Zeneth is going strongly – expansion of functionality – sustained growth of the knowledge base

• Collaborative group of members – co-direction of development – contributions – handling of confidential data

• Benefits of participation – faster identification of degradation pathways – preserving knowledge & training/teaching – sustained development and support

William Button

Alex Cayley

Tony Long

Nicole McSweeney

Ernest Murray

Rob Toy

Thanks to …

Steve Baertschi Eli Lilly

Rhonda Jackson

J&J

Mark Kleinman GSK

Darren Reid

Amgen