Data analysis and Visualisation Techniques for Compound Combination Modelling

26
Data Analysis and Visualization Techniques for Compound Combination Modelling Cambridge Cheminformatics Network Meeting, 25/11/2015 Richard Lewis Centre for Molecular Informatics

Transcript of Data analysis and Visualisation Techniques for Compound Combination Modelling

Page 1: Data analysis and Visualisation Techniques for Compound Combination Modelling

Data Analysis and Visualization Techniques for Compound Combination Modelling

Cambridge Cheminformatics Network Meeting, 25/11/2015

Richard Lewis

Centre for Molecular Informatics

Page 2: Data analysis and Visualisation Techniques for Compound Combination Modelling

Introduction

• Briefly introduce compound combinations and synergy

• Detail experimental setups and measurement techniques of synergy on a per combination (micro scale) basis.

• Explain the concerns for the design of different macro scale combination experiments.

• Introduce Synergy Maps, an improvement over currently used visualization techniques for large combination datasets.

Page 3: Data analysis and Visualisation Techniques for Compound Combination Modelling

A very brief background: Compound Combinations

• Two or more pure compounds ‘mixed together’.

• Offer the potential for improved pharmaceutical treatment options:

• Increased efficacy

• Increased selectivity

• Reduced toxicity

• Reduced chance of side effects.

• Especially important for complex disease areas such as cancer and neurodegenerative diseases.

Review: Bulusu KC et al.: Modelling of compound combination effects and applications to efficacy and toxicity: state-of-the-art, challenges and perspectives. Drug Discovery Today 2015.

Page 4: Data analysis and Visualisation Techniques for Compound Combination Modelling

Synergy

“The phenomenon of super-additivity of the therapeutic effect of a combination.”

Synergy

Antagonism• Two compounds, when applied together, cause a greater response than expected based on their individual application.

• Easily seen from dose response surfaces.

• How do you measure synergy?

Bulusu KC et al.: Modelling of compound combination effects and applications to efficacy and toxicity: state-of-the-art, challenges and perspectives. Drug Discovery Today 2015.

Page 5: Data analysis and Visualisation Techniques for Compound Combination Modelling

Experimental Design (micro scale)

Checkerboard Fixed dose ratios

(synthetic data)(synthetic data)

Page 6: Data analysis and Visualisation Techniques for Compound Combination Modelling

Additivity Models for Measuring Synergy

Additivity models predict the response of a combination as a function of the responses of its constituent compounds, assuming they don’t interact.

Therefore, the degree of synergy exhibited by the combination may be interpreted as the deviation of observed response from that expected from an additivity model.

There are many different models, with varying names in the literature. Some examples:

• Loewe Additivity

• Bliss Independence

• More Recent: SANE [1], Zero Interaction Potency [2]

[1] Jodrell group, http://www.cruk.cam.ac.uk/research-groups/jodrell-group/combenefit, unpublished[2] Yadav et al.: Searching for drug synergy in complex dose-response landscapes using an interaction potency model. Computational and Structural Biotechnology Journal 2015.

Page 7: Data analysis and Visualisation Techniques for Compound Combination Modelling

An Example: Bliss Independence Model

• The Bliss Independence model assumes compounds act independently (i.e. through different mechanisms).

• Uses single agent response curves to model a prediction surface.

• The deviations of the observed combination data from the predictions gives a measure of synergy.

Page 8: Data analysis and Visualisation Techniques for Compound Combination Modelling

An Example: Bliss Independence Model

• For a cytotoxicity response experiment, 100% of cells are alive before the combination is applied.

100%

Page 9: Data analysis and Visualisation Techniques for Compound Combination Modelling

An Example: Bliss Independence Model

• Compound A is applied at concentration a. [A] = a has a known response from single agent data of 0.4 (i.e. kills 40% of cells).

60%40% Compound A

Page 10: Data analysis and Visualisation Techniques for Compound Combination Modelling

An Example: Bliss Independence Model

• Compound B is applied at concentration b. This also has a known response from single agent data of 0.3 (i.e. kills 30% of cells). Assuming A and B operate independently, this will kill 30% of cells remaining from the application of compound A.

60%40%

70%30%

Compound A

Compound B

Page 11: Data analysis and Visualisation Techniques for Compound Combination Modelling

An Example: Bliss Independence Model

• Therefore, the bliss independence prediction for the combination is 42% surviving, or a response of 0.58.

60%40%

70%30%

42%58%

Compound A

Compound B

Total Combination

Page 12: Data analysis and Visualisation Techniques for Compound Combination Modelling

Bliss Independence: Example workflow

Take single agent data

Predict surface

Calculate

differenceAggregate for single

score

16.59

Page 13: Data analysis and Visualisation Techniques for Compound Combination Modelling

Combination Screen (macro scale data) Experimental Design

Specific

Row Matrix (one vs all)

Sparse Matrix (some vs some)

Full Matrix (all vs all)

General

• Combination screening is expensive

• The choice of which combinations to test is important

Page 14: Data analysis and Visualisation Techniques for Compound Combination Modelling

All vs All Example

NCATS Malaria Dataset

• 56 antibiotic and antimalarial compounds

• Every pairwise combination (1540 combinations)

• Tested against lysed red blood cells infected with P. falciparum parasite.

• For 3 malarial strains: 3D7, HB3, DD2

Mott BT et al.: High-throughput matrix screening identifies synergistic and antagonistic antimalarial drug combinations, Sci. Rep. 5, 13891

Page 15: Data analysis and Visualisation Techniques for Compound Combination Modelling

All vs All Example

NCATS Malaria Dataset

• 56 antibiotic and antimalarial compounds

• All (1540) pairwise combinations

• Tested against lysed red blood cells infected with P. falciparum parasite.

• For 3 malarial strains: 3D7, HB3, DD2

• Multiple metrics measured, pGamma found to be most reliable metric.

• How to visualize?

Mott BT et al.: High-throughput matrix screening identifies synergistic and antagonistic antimalarial drug combinations, Sci. Rep. 5, 13891

Page 16: Data analysis and Visualisation Techniques for Compound Combination Modelling

Traditional Visualizations for Combination Datasets

Synergy Heatmap

 Lewis R et al.: Synergy Maps: exploring compound combinations using network-based visualization. J Cheminform 2015, 7:36.

Page 17: Data analysis and Visualisation Techniques for Compound Combination Modelling

Traditional Visualizations for Combination Datasets

Synergy Network

 Lewis R et al.: Synergy Maps: exploring compound combinations using network-based visualization. J Cheminform 2015, 7:36.

Page 18: Data analysis and Visualisation Techniques for Compound Combination Modelling

Problems with Traditional Visualizations

• Crowded with data.

• Problems with scaling to larger data sets.

• Difficult to deduce the relationship between synergistic mechanisms and chemical structure.

• How is chemical space usually visualized?

Page 19: Data analysis and Visualisation Techniques for Compound Combination Modelling

PCA of Structural Space

Page 20: Data analysis and Visualisation Techniques for Compound Combination Modelling

t-SNE of Biological Space

Mervin LH et al.: Target prediction utilising negative bioactivity data covering large chemical space. J. Chem. Inf. 2015

Page 21: Data analysis and Visualisation Techniques for Compound Combination Modelling

Synergy Maps

Page 22: Data analysis and Visualisation Techniques for Compound Combination Modelling

Synergy Maps

HDAC InhibitorsPI3K/mTOR

Inhibitors

Antibiotics

Antimalarials

Artemisinin Antimalarials

Page 23: Data analysis and Visualisation Techniques for Compound Combination Modelling

Synergy Maps: An interactive web app

richlewis42.github.io/synergy-maps

JavaScript D3.js AngularJS

 Lewis R et al.: Synergy Maps: exploring compound combinations using network-based

visualization. J Cheminform 2015, 7:36. DOI: 10.1186/s13321-015-0090-6

Page 24: Data analysis and Visualisation Techniques for Compound Combination Modelling

Conclusions

• Compound combinations is an rapidly moving field with many potential medicinal benefits (amongst others).

• Measuring synergy for a given combination is a non-trivial task

• The design of combination screens is challenging

• These datasets are multifaceted, thus their visualization is also hard

• Synergy maps is a potential solution for an all-in-one initial visualization tool for a combination dataset.

Page 25: Data analysis and Visualisation Techniques for Compound Combination Modelling

Future Work

• Expanding the server side to accept raw surfaces

• Implement many different synergy metrics

• Allow for the upload of datasets

• Implement more advanced export of static synergy maps

Page 26: Data analysis and Visualisation Techniques for Compound Combination Modelling

Acknowledgements

Andreas Bender

Mixture Modelling Group

Tamás Korcsmaros Rajarshi Guha (NCATS)Murat CokolEugene MuratovKrishna BulusuDan MasonRanjoo ChoiYasaman KalandarMotamediSiti Mohamad-ZobirAzedine Zoufir

Bender Group

Funding: