36
Practical Applications of Matched Molecular Pairs at Vernalis Steve Roughley Richard Sherhod What are they? How do we find them? How to deploy for users?

Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

  • Upload
    others

  • View
    2

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

Practical Applications of Matched Molecular Pairs at Vernalis

Steve RoughleyRichard Sherhod

• What are they?• How do we find them?• How to deploy for users?

Page 2: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

2 16 March 2017

About Vernalis

• Expertise• Fragments and structure‐based drug discovery(Protein Science, Structural Biology, Chemistry)

• Therapeutic areas• Oncology, CNS, infectious diseases

• Location• Based in Granta Park, outside Cambridge, UK

Trusted community contributor since 2013 (2 KNIME‐trained developers)

Page 3: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

3 16 March 2017

Matched Molecular Pairs (MMPs)

“MMP can be defined as a pair of molecules that differ in only a minor single point change”

(Wikipedia)

•Multiple open‐source implementations in various forms•At least 2 in KNIME

• Vernalis• Erlwood

• Recently reviewed• Christian Tyrchan and Emma Evertsson, Comput. & Struct. Biotech. J., 2017, 15, 86‐90

Definition

CHEMBL2263252 CHEMBL60592

Page 4: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

4 16 March 2017

Anatomy of a Matched Molecular Pair

• Hussain‐Rea Algorithm3

• Identify bonds that can be broken• Eg acyclic bonds

• Break molecule along each matching bond in turn• Match identical ‘Keys’

• ‘Values’ form pair transforms

*

*

**

identical keysValues form Pair Transform

cutMolecule A

Molecule B

3. Jameed Hussain, Ceara Rea, J. Chem. Inf. Model., 2010, 50, 339–348

Page 5: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

5 16 March 2017

Anatomy of a Matched Molecular Pair

• Hussain‐Rea Algorithm3

• Identify bonds that can be broken• Eg acyclic bonds

• Break molecule along each matching bond in turn• Match identical ‘Keys’

• ‘Values’ form pair transforms

identical keysValues form Pair Transform

cut

3. Jameed Hussain, Ceara Rea, J. Chem. Inf. Model., 2010, 50, 339–348

CHEMBL2263252

CHEMBL60592

Page 6: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

6 16 March 2017

Multi‐cut pairs

• Pairs also can be formed by cutting 2 or more bonds simultaneously• Allows scaffold replacement transforms• Need to track which breaking bond is which

Identical KeysValues form Pair Transform

Molecule A

Molecule A

2*1*

1* 2*

1* 2*2*

1*

cut

Page 7: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

7 16 March 2017

Multi‐cut pairs

• Pairs also can be formed by cutting 2 or more bonds simultaneously• Allows scaffold replacement transforms• Need to track which breaking bond is which

Values form Pair Transform

cutCHEMBL373838

CHEMBL1368873

Identical Keys

Page 8: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

8 16 March 2017

Transforms application

Original molecules interconvert when transform is applied

A>>B

Molecule A Molecule B

CHEMBL2263252 CHEMBL60592

A>>B

**‘A’ ‘B’

A>>B

CHEMBL1350874Not found in ChEMBL

Other molecules generate new ideas

Transform takes no account of relevance or context

Page 9: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

9 16 March 2017

DESTRUCTION TESTING

Can we fragment all of ChEMBL?A ‘reasonable’ ‘representative’ test set

Page 10: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

10 16 March 2017

Pre‐processing – “Speedy SMILES”

• Fast String‐based SMILES string manipulation•No chemical toolkit conversion

e.g. c1cc[nH]c1C(=O)OCCN(C)C

• Streamable• Example application – pre‐processing ChEMBL

• De‐salt• Remove large (HAC>40) or small (HAC<8) molecules• and those with a net charge or large number of charges

Vernalis Community Nodes

Page 11: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

11 16 March 2017

SpeedySMILES pre‐processingChEMBL

1,581,653 molecules inProcessed 76 seconds1,486,077 molecules out

Failure Category Count

HAC < 8 or > 50 63,920

Non‐neutral 30,706

Total Charges > 4 949

Broken Bonds 1

Page 12: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

12 16 March 2017

ChEMBL fragmentation

• 1,420,462molecules fragmented• 1‐10 cuts• Non‐functional group single bonds• Maximum 10,000 fragmentations / molecule

• 134,020,007 fragments• 139,679 failed rows:

• 10h30min (Intel® Core™ i7‐4770 @ 3.4GHz; W10)• 10 threads; 500 rows buffer• ‐Xmx16329m

The numbers…

Failure category Count

Complexity limit 139,550

Too few matching bonds 79

Valence error 25

Kekulisation error 25

This version will be released to the community ‘imminently’

Page 13: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

13 16 March 2017

Example failure rows

CHEMBL2297882Molecule failed complexity limit15965 possible fragmentations

CHEMBL1698868No matching bounds found, or too few to cut

CHEMBL178180Error parsing … Explicit valence for atom # 8 Te, 4, is greater than permitted

CHEMBL3188982Error parsing… OC(=O)C(=O)Nc1cccc(c1)c2nnnn2 …Unkekulized atoms

CHEMBL2006679Error parsing … Unkekulized atoms 4

SP

HN NH

P

NHHN

HN P

N

PNN

N

NN

O

New Vernalis Matching Bonds Renderer Node

+H

Page 14: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

14 16 March 2017

Matched Molecular Pairs (MMPs)

•MMP Concept has been extended to improve utility• Data analysis

• What effect does a transform have on data, e.g. Metabolism/Stability, hERG binding, target binding?

• Matched Molecular Series• When my series is seen in activity order, what other new members are commonly ‘better’?

• Fingerprint similarity• How closely related is the surrounding chemical matter to my input molecule?

• 3D Matched Pairs• Molecular shape/pharmacophore presentation

• In all cases, provides additional ‘context’ to the pairs

• Can be used for substituent analysis/replacement or scaffold replacement 

Page 15: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

MMAnalyser: Applied MMP/S Analysis

Richard [email protected]

Page 16: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

16 16 March 2017

MMAnalyser

• KNIME Web Portal application• Composed of multiple interactive KNIME workflows

•Allows chemists to do matched‐molecular pair/series analysis

• Guides users through the analysis process

Page 17: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

17 16 March 2017

MMAnalyser: Application

• Two interactive workflows for chemists• MMPair Analyser – matched‐molecular pair analysis• MMSeries Analyser – matched molecular series analysis

• Interactive admin workflows for database maintenance• Rebuilding MMPair and MMSeries databases from pre‐defined sources

Kenny, P.W. & Sadowski, J., 2005. Structure Modification in Chemical Databases. In Wiley‐VCH Verlag GmbH & Co. KGaA, pp. 271–285. Available at: http://doi.wiley.com/10.1002/3527603743.ch11 [Accessed March 6, 2017].

Wawer, M. & Bajorath, J., 2011. Local Structural Changes, Global Data Views: Graphical Substructure−Ac vity Rela onship Trailing. Journal of Medicinal Chemistry, 54(8), pp.2944–2951. Available at: http://pubs.acs.org/doi/abs/10.1021/jm200026b [Accessed March 6, 2017].

Page 18: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

18 16 March 2017

MMPair Analyser: Application

Pre‐generated transformations with observation data

Input structure

MMP Analysis• All transforms applied to the input structure

• Results filtered and sorted by:• Observation count• Enrichment of positive observations

Page 19: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

19 16 March 2017

MMPair Analyser: Database

ChEMBL data• Molecule dictionary• Compound structures• Compound properties (QED)• Compound records• Activities• Assays• Documents

Observations

Filtered structures

Generate MMPs1. Fragment structures

• Fragments (values) and scaffolds (keys)2. Add hydrogens to fragments and scaffolds3. Generate transformations from fragments

• Record ID of left and right structures

Gather evidence1. Filter transformations by observation count2. Get observed changes in property for each 

transformation3. Calculate enrichment of positive observations4. Perform one‐tailed binomial test

• Keep transforms with p >= 0.05

Page 20: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

20 16 March 2017

MMPair Analyser: Demo

Page 21: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

21 16 March 2017

MMSeries Analyser

• Extension of MMPs to three or more R‐groups (values)• Originally proposed by Waver & Bajorath (2011)• Several implementations, e.g. MATSY (O’Boyle et al. 2014)

• R‐groups ordered by the properties of their parent

Ms

MeOR

pIC50 R pIC50

7.00 H 8.30

7.68 F 8.00

8.51 Cl 7.77

8.77 Br 8.89Br > Cl > F > H Br > H > F > Cl

O’Boyle, N.M. et al., 2014. Using Matched Molecular Series as a Predictive Tool To Optimize Biological Activity. Journal of Medicinal Chemistry, 57(6), pp.2704–2713. Available at: http://pubs.acs.org/doi/abs/10.1021/jm500022q [Accessed March 3, 2017].

Wawer, M. & Bajorath, J., 2011. Local Structural Changes, Global Data Views: Graphical Substructure−Ac vity Rela onship Trailing. Journal of Medicinal Chemistry, 54(8), pp.2944–2951. Available at: http://pubs.acs.org/doi/abs/10.1021/jm200026b [Accessed March 6, 2017].

Matched‐molecular Series Analysis

Page 22: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

22 16 March 2017

MMSeries Analyser: Application

Input structuresWith unique IDs and 

numeric data

MMS Analysis• Query structures are fragmented into scaffolds and R‐groups• Sets of R‐groups, their scaffold and data are arranged into series

• Query series are compared to pre‐generated series• Common R‐groups are recorded

• Query and database series ordered by data• Spearman's rank correlation between matching series calculated• Matching series sorted by rank correlationPre‐generated sets of scaffolds 

and R‐groups with data

Page 23: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

23 16 March 2017

MMSeries Analyser: Application

Scaffold Rank Correlation

86 69 63 60 59 52 46 34

0.975 100 100 95 94 92

0.9 72 61 16 33 7

0.5 95.6 85.2 95.7 90 78.9

Page 24: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

24 16 March 2017

MMSeries Analyser: Database

ChEMBL data• Molecule dictionary• Compound structures• Compound properties (MW)• Compound records• Activities• Assays• Documents

Observations

Filtered structures

Generate MMPs1. Fragment structures into R‐groups and scaffolds

• 4 methods including Hussein/Rea rules2. Group R‐groups into series by common parent 

scaffolds3. Keep series of 3 or more R‐groups4. Record IDs of parent structures

Gather evidenceAssociate parent structures with observation data

Page 25: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

25 16 March 2017

MMSeries Analyser: Demo

Page 26: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

26 16 March 2017

MMAnalyser: Possible Improvements

• Present observation data to the user• Direct the user to relevant source material

•More datasets• Better (more robust) processing

• Incorporate transformation site similarity (MMPs)• Associate transformations with fingerprints from their parent scaffold(s)

• Incorporate scaffold similarity (MMSs)• Filter/order series by similarity to query scaffold

Page 27: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

27 16 March 2017

Acknowledgments

•Vernalis colleagues

•Greg Landrum (RDKit)

• ChEMBL

• KNIME

Matched‐molecular series implementation:

Hunt, P. et al., 2017. Practical applications of matched series analysis: SAR transfer, binding mode suggestion and data point validation. Future Medicinal Chemistry, 9(2), pp.153–168. Available at: http://www.future‐science.com/doi/10.4155/fmc‐2016‐0203 [Accessed March 3, 2017].

Page 28: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

Thank you!

Page 29: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

29 16 March 2017

BACKUP SLIDES

Page 30: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

31 16 March 2017

Stereochemistry

• Fragmentation can create new chiral centres / double bond geometries• Existing absolute/unknown must be preserved

Known Stereocentre?

Yes No

Unk

nown/Ra

cemic

Stereo

centre? Yes

No

Page 31: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

32 16 March 2017

Stereochemistry

• Fragmentation can create new chiral centres / double bond geometries• Existing absolute/unknown must be preserved

Known Stereocentre?

Yes No

Unk

nown/Ra

cemic

Stereo

centre? Yes

No

“KNOWN KNOWNS”We know we know about stereochemistry

Page 32: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

33 16 March 2017

Stereochemistry

• Fragmentation can create new chiral centres / double bond geometries• Existing absolute/unknown must be preserved

Known Stereocentre?

Yes No

Unk

nown/Ra

cemic

Stereo

centre? Yes

No

“KNOWN UNKNOWN”We know unknown or racemic

Page 33: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

34 16 March 2017

Stereochemistry

• Fragmentation can create new chiral centres / double bond geometries• Existing absolute/unknown must be preserved

Known Stereocentre?

Yes No

Unk

nown/Ra

cemic

Stereo

centre? Yes

No ?“UNKNOWN UNKNOWN”We have no idea…

Page 34: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

35 16 March 2017

Stereochemistry

• Fragmentation can create new chiral centres / double bond geometries• Existing absolute/unknown must be preserved

Known Stereocentre?

Yes No

Unk

nown/Ra

cemic

Stereo

centre? Yes

No ? or

Page 35: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

36 16 March 2017

Sneak Preview – Upcoming revised release

•Memory leak fixed• Survives ‘destruction testing’•New fragmentation type added

• More flexibility for custom types•New parallelised pair generation nodes• Transform filtering options• Reference table version

• Only generates pairs between rows from the two tables

• Improved Rendering/Filtering nodes

Page 36: Practical Applications of Matched Molecular Pairs at Vernalis · •ChEMBL •KNIME Matched‐molecular series implementation: Hunt, P. et al., 2017. Practical applications of matched

37 16 March 2017

Attachment point fingerprints

• RDKit Morgan ECFP‐like fingerprint• Rooted at the attachment point atom for each ‘key’ component• Default radius 4, size 2048 bit• Calculated during fragmentation in Vernalis Nodes

1:  10000000000100000000010100000100

1A

O

1:  00001000000010000000100000000101

1:  000000000000100000000001000001002:  10000000000001000000001100000100

Example 32‐bit AP fingerprints