1

Chemoinformaticswith artificial intelligence

Péter Antal

Computational Biomedicine (Combine) workgroupDepartment of Measurement and Information Systems,

Budapest University of Technology and Economics

Overview

• Chemoinformatics

• Artificial intelligence/machine learning

• The data flood in life sciences

• The data and knowledge fusion challenge

• The semantic unification in chemoinformatics

• Artificial intelligence in drug discovery

• Examples:

– Drug repositioning

– Drug-target interaction prediction

2

Chemoinformatics

• Gasteiger, Johann, and Thomas Engel, eds.

Chemoinformatics: a textbook. John Wiley & Sons, 2006.

• Bajorath, Jürgen. Chemoinformatics for Drug Discovery.

John Wiley & Sons, 2013.

• Karthikeyan, Muthukumarasamy, and Renu Vyas.

Practical Chemoinformatics. Springer, 2014.

• Brown, Nathan. In Silico Medicinal Chemistry:

Computational Methods to Support Drug Design. No.

8. Royal Society of Chemistry, 2015.

3

MK&RV:Practical chemoinformatics

4

1. Open-Source Tools, Techniques, and Data in Chemoinformatics

Semantic(-web) technologies

2. Chemoinformatics Approach for the Design & Screening of Focused Virtual Libraries

Prioritization methods using data and knowledge fusion

3. Machine Learning Methods in Chemoinformatics for Drug Discovery

Prediction methods using data and knowledge fusion

4. Docking and Pharmacophore Modelling for Virtual Screening

Priors from docking

5. Active Site-Directed Pose Prediction Programs for Efficient Filtering of Molecules

Binding sites, pockets and latent dimensions

6. Representation, Fingerprinting, and Modelling of Chemical Reactions

7. Predictive Methods for Organic Spectral Data Simulation

8. Chemical Text Mining for Lead Discovery

9. Integration of Automated Workflow in Chemoinformatics for Drug Discovery

Visual data analytics and workflow systems

10.Cloud Computing Infrastructure Development for Chemoinformatics

Karthikeyan, Muthukumarasamy, and Renu Vyas. Practical Chemoinformatics.

Springer, 2014.

Automating

drug discovery

Schneider, Gisbert. "Automating drug discovery." Nature Reviews Drug Discovery 17.2 (2018): 97.

Design cycle

Automated drug discovery facility

Active learning with microfluidics

Artificial intelligence

• IBM Grand Challenge

– 1997: Deep Blue wins human champion G.

Kasparov.

– 1999-2006<: Blue Gene, protein prediction

– 2011: Watson

• Natural language processing

• inference

• Game theory

IBM Watson (2011): Jeopardy

Clinical decision support systems

Watson for Oncology – assessment and advice cycle

www.avanteoconsulting.com/machine-learning-accelerates-cancer-research-discovery-innovation/

• Google DeepMind

• Monte Carlo tree

search

• 2016: 9 dan

• 2017: wins against

human champion

Go:

• 2017: Carnegie Mellon University MI:

Libratius

• Pittsburgh Supercomputing Center:

– 1.35 petaflops computation

– 274 Terabytes memory

Poker: Libratus

• Teaching + Learning: learning from manual

and from practice

Machines playing Civilization

Proportion of wins

Playing computer games

• YOLO (you only look once)

Vision: YOLO

https://www.ted.com/talks/joseph_redmon_how_a_computer_learns_to_recognize_obj

ects_instantly#t-409586

Emotion detection, sentiment analysis

https://www.ted.com/talks/rana_el_kaliouby_this_app_knows_how_you_feel_fro

m_the_look_on_your_face

Walking, movements

Real-time translation

D.Adams: Hitchhiker's Guide to the Galaxy"Pilot Translating Earpiece

• ~„big data failed, AI correctly predicted

the upset victory” (correct prediction of

election in the US 3 times in a row)

Political analytics: MogIA

Automated essay scoring (AES)

• Juridical decisions:

– Human experts: 66% identical decision.

– Katz, D.M., Bommarito II, M.J. and Blackman,

J., 2017. A general approach for predicting

the behavior of the Supreme Court of the

United States. PloS one, 12(4), p.e0174698.

• 1816-2015 esetek

• 70%< accuracy

– COMPAS CORE

Legal applications of AI

February 28, 2019 21

http://beauty.ai/• A beauty contest was judged by AI and the robots

didn't like dark skin, Guardian

• Another AI Robot Turned Racist, This Time At Beauty

Contest, Unilad

Beauty.AI

February 28, 2019 22

• Turing-test, Loebner-prize

• Tay was an artificial intelligence chatterbot released by

Microsoft Corporation on March 23, 2016. Tay caused

controversy on Twitter by releasing inflammatory tweets

and it was taken offline around 16 hours after its launch.[1]

Tay was accidentally reactivated on March 30, 2016, and

then quickly taken offline again.

Chatbot: Tay

• Gatys, L.A., Ecker,

A.S. and Bethge,

M., 2015. A neural

algorithm of artistic

style. arXiv

preprint

arXiv:1508.06576.

Reproduction of artistic style

Automated discovery systems Langley, P. (1978). Bacon: A general discovery system. Proceedings of the Second Biennial Conference of the Canadian Society for Computational Studies of Intelligence (pp. 173-180). Toronto, Ontario.

…

Chrisman, L., Langley, P., & Bay, S. (2003). Incorporating biological knowledge into evaluation of causal regulatory hypotheses. Proceedings of the Pacific Symposium on Biocomputing (pp. 128-139). Lihue, Hawaii.

(Gene prioritization…)

R.D.King et al.: The Automation of Science, Science, 2009

„Machine science”Swanson, Don R. "Fish oil, Raynaud's syndrome, and undiscovered public knowledge." Perspectives in biology and medicine 30.1 (1986): 7-18.

Smalheiser, Neil R., and Don R. Swanson. "Using ARROWSMITH: a computer-assisted approach to formulating and assessing scientific hypotheses." Computer methods and programs in biomedicine 57.3 (1998): 149-153.

D. R. Swanson et al.: An interactive system for finding complementary literatures: a stimulus to scientific discovery, Artificial Intelligence, 1997

James Evans and Andrey Rzhetsky: Machine science, Science, 2013

„Soon, computers could generate many useful hypotheses with little help from

humans.”

State of the art

26

• Deep Blue defeated the reigning world chess champion Garry Kasparov in 1997

• Proved a mathematical conjecture (Robbins conjecture) unsolved for decades

• No hands across America (driving autonomously 98% of the time from Pittsburgh to San Diego)

• During the 1991 Gulf War, US forces deployed an AI logistics planning and scheduling program that involved up to 50,000 vehicles, cargo, and people

• NASA's on-board autonomous planning program controlled the scheduling of operations for a spacecraft

• Proverb solves crossword puzzles better than most humans

• Google search

• Object recognition…

Hallmarks of a new AI era?

27

Factors behind the „A.I./learning

hype”• New theory?

– Unified theory of AI?

– A new machine learning approach?

• New hardware? (computing power..)

– Graphics cards (GPUs)?

– Quantum computers?

• New resources?

– Data?

– Knowledge?

– Money?

– Brains/Minds?

Milestones and phases in AI• ~1930: Zuse, Neumann, Turing..: „instruction is data”:

– Laws of nature can be represented, „executed”/simulated with modifications, learnt

– Knowledge analogously: representation, execution, adaptation and learning

• 1943 McCulloch & Pitts: Boolean circuit model of brain

• 1950 Turing's "Computing Machinery and Intelligence"

• 1956 Dartmouth meeting: the term "Artificial Intelligence”

• 1950s Early AI programs (e.g. Newell & Simon's Logic Theorist)

• The psysical symbol system hypothesis: search

• 1965 Robinson's complete algorithm for logical reasoning

• 1966—73 AI discovers computational complexityNeural network research almost disappears

• 1969—79 Early development of knowledge-based systems

• The knowledge system hypothesis: knowledge is power

• 1986-- Neural networks return to popularity

• 1988-- Probabilistic expert systems

• 1995-- Causality research

• The „big data” hypothesis: let data speak

• 1995-- Emergence of machine learning

• 2005/2015-- Emergence of autonomous adaptive decision systems („robots”, agents)

• The autonomy hypothesis??

Computational

complexity

Knowledge

representation

Exp

ert

syste

ms

Thresholds of

knowledge

Ma

ch

ine

lea

rnig

Statistical

complexity

Ad

ap

tive

de

cis

ion

syste

ms

Computer

Reminder:

Automating

drug discovery

Schneider, Gisbert. "Automating drug discovery." Nature Reviews Drug Discovery 17.2 (2018): 97.

Design cycle

Automated drug discovery facility

Active learning with microfluidics

The data flood in life sciences

Heterogeneous data in biomedicine

Genome(s)

Phenome (disease, side effect)

Transcriptome

Proteome

Metabolome

Environment&life style

Drugs

Moore’s Law for Data Explosion (Carlson’s law)

Sequencing

costs per mill.

base

Publicly

available

genetic data

NATURE, Vol 464, April 2010

• x10 every 2-3 years

• Data volumes and

complexity that IT has

never faced before…

Bioactivity databases I.

34

•Targets: 10,774

•Compound records: 1,715,667

•Distinct compounds: 1,463,270

•Activities: 13,520,737

•Publications: 59,610

ChEMBL is a database of bioactive drug-like small molecules, it contains 2-D

structures, calculated properties (e.g. logP, Molecular Weight, Lipinski

Parameters, etc.) and abstracted bioactivities (e.g. binding constants,

pharmacology and ADMET data).

https://www.ebi.ac.uk/chembl

Bioactivity databases II.

Compounds: 97,127,348

Substances: 252,300,917

BioAssays: 1,067,565

Tested Compounds: 3,417,415

Tested Substances: 5,591,261

RNAi BioAssays: 173

BioActivities: 239,680,570

Protein Targets: 12,159

Gene Targets: 58,18635

Bioactivity databases III:ExCAPE-DB

36

Sun, J., Jeliazkova, N., Chupakhin, V., Golib-Dzib, J.F., Engkvist, O.,

Carlsson, L., Wegner, J., Ceulemans, H., Georgiev, I., Jeliazkov, V. and

Kochev, N., 2017. ExCAPE-DB: an integrated large scale dataset

facilitating Big Data analysis in chemogenomics. Journal of

cheminformatics, 9(1), p.17.

Data: chemogenomics screening

• Justin Lamb: The Connectivity Map: a new tool for

biomedical research, Nature, 7,pp 54-60, 2007

Compounds Cell lines

Each cell is

transcriptional

proifle

Repositories for gene expression

• Gene Expression Omnibus (NCBI)

• http://www.ncbi.nlm.nih.gov/geo/

STRING - Protein-Protein Interactions

• http://string-db.org/

Number of genome-wide association studiesTota

l N

um

ber

of

Public

ations

Calendar Quarter

0

200

400

600

800

1000

1200

1400

2005 2006 2007 2008 2009 2010 2011 2012

1350

NHGRI GWA Catalog

www.genome.gov/GWAStudie

s

www.ebi.ac.uk/fgpt/gwas/

Published Genome-Wide Associations through 12/2012

Published GWA at p≤5X10-8 for 17 trait categories

Genetic overlap based disease maps

L.A.Barabási:PNAS, 2007, The human disease network

Epidemiologocal disease maps

Marx, P., Antal, P., Bolgar, B., Bagdy, G., Deakin, B. and Juhasz, G., 2017. Comorbidities in the diseasome are

more apparent than real: What Bayesian filtering reveals about the comorbidities of depression. PLoS

computational biology, 13(6), p.e1005487.

Number of biomedical publications

44

Little Science, Big Science, by

Derek J. de Solla Price, 1963

0

200000

400000

600000

800000

1000000

1200000

1950 1960 1970 1980 1990 2000 2010

Number of annual papers

Unification of biology: Gene Ontology

• Ontologies:

– Gene Ontology (GO): http://www.geneontology.org/

– Enzyme Classification (EC)

– Unified Medical Language Systems (UMLS)

– OBO

The Human Phenotype Ontology

http://human-phenotype-ontology.github.io/

Semantic publishing:

papers vs DBs/KBs

M. Gerstein, "E-publishing on the Web: Promises, pitfalls, and payoffs for bioinformatics," Bioinformatics, 1999

M. Gerstein: Blurring the boundaries between scientific 'papers' and biological databases, Nature, 2001

P. Bourne, "Will a biological database be different from a biological journal?," Plos Computational Biology, 2005

M. Gerstein et al: "Structured digital abstract makes text mining easy," Nature, 2007.

M. Seringhaus et al: "Publishing perishing? Towards tomorrow's information architecture," Bmc Bioinformatics,

2007.

M. Seringhaus: "Manually structured digital abstracts: A scaffold for automatic text mining," Febs Letters, 2008.

D. Shotton: "Semantic publishing: the coming revolution in scientific journal publishing," Learned Publishing, 2009

47

The fusion challenge

in drug discovery

Combination of

elements

geneg

en

e

target

co

mp

ou

nd

ge

ne

disease

binding site

co

mp

ou

nd

target protein

bin

din

g s

ite

product

ge

ne

gene

TF

BS

pathway

ge

ne

disease

pa

thw

ay

transcription factor

binding site

pro

du

ct

AT

C

GO

EC

HPO

E D. Green et al. Nature 470, 204-213 (2011) doi:10.1038/nature09764

Accomplishments of genomics research

Pharma productivity (~gap)

Mullard, A., 2017. 2016 FDA drug approvals. Nature Reviews Drug Discovery,

16(2), pp.73-76.

The fusion bottleneck

(~limits of personal cognition)

Watson?

The Science Behind an Answer

• http://www-03.ibm.com/innovation/us/watson/what-is-watson/science-behind-an-answer.html

Network of databases in 2000

54

• 10k< relevant biological

databases and knowledge-bases

• Petabytes of sequence and

high-throughput gene/protein

data

• ~10.000.000 concepts and

relations explicitly in

knowledge bases

Linked Open Data in 2017

55

Linking Open Data cloud diagram 2017, by Andrejs Abele, John P. McCrae, Paul Buitelaar, Anja Jentzsch and Richard

Cyganiak. http://lod-cloud.net/

Approaches to fusion

• Encyclopedists:

– Wikipedia, Wikidata,

– Linked Open Data (LOD),

– Semantic unification

• Automated cross-domain querying

– Forms

– Workflow systems

– Natural language understanding, Machine reading

• Automated reasoning

– Watson

• Automated discovery systems („Automation of science”)

– Adam, Eve

• Large-scale similarity-based fusion applied in repositioning

56

Semantic unification in

chemoinformatics

Semantic Web

• Tim Berners-Lee, 1999, „I have a dream...”, W3C

• Web of data, Web 3.0

• Share, reuse, querying, integration of data, automatic processing, reasoning

• Publishing data in human readable HTML documents to machine readable documents

• Linked Data

The Internet network: nodes are computers or post-pc devices and links are wired or

wireless connections between them.

https://users.dimi.uniud.it/~massimo.franceschet/netart/talk/netart.html

The Resource Description Framework

(RDF)

• The data model of the Semantic Web

• RDF statement

– subject: resource identified by an IRI

– predicate (property): resource identified by an

IRI

– object: resource or literal (constant value)

• Graph databases of RDF triples

60

Relational databases vs.

Triplestores (graph databases)Relational databases• Relations are separated from data (cases)

• Tables&keys define the formal model (syntax)

for the data (cases)

• Model-based (~predefined)

• Meaning (semantics) is informal (out of scope

of the DB)

• Singular databases (~they are separated)

Triplestores• Unified representation of relations and data

• Triples („graph database”) stores the dynamic

model for the data, together with the factual

data

• Model-free (~relations as data)

• Meaning is defined by the (explicit) relations

(~ontology)

• Linked open data space (using universal

identifiers & ontologies)

61

Semantic technologies for drug

discovery

• Whitaker, B.J. and Rzepa, H.S., 1995. Chemical publishing via the

Internet. In International chemical information conference (pp. 62-71).

• Murray-Rust, P., Rzepa, H.S., Wright, M. and Zara, S., 2000. A

universal approach to web-based chemistry using XML and CML.

Chemical Communications, (16), pp.1471-1472.

• Murray-Rust, P. and Rzepa, H.S., 2002. Scientific publications in

XML-towards a global knowledge base. Data Science Journal, 1,

pp.84-98.

• Murray-Rust, P., 2008. Chemistry for everyone. Nature, 451(7179),

pp.648-651.

62

A problem with public data: parallel works on cleaning...integration

63

• Discovery Platform for cross-domain fusion.

• Public, curated, linked data.

– The data sources you already use, integrated and

linked together: compounds, targets, pathways,

diseases and tissues.

• Everything in triples: Subject-predicate-object

64

Open Pharmacological Space

Precursor: Gene Ontology: tool for the unification of biology, Nature, 2000

@gray_alasdair Big Data Integration 65

• Discovery Platform to cross barriers.

• The data sources you already use, integrated

and linked together: compounds, targets,

pathways, diseases and tissues.

• ChEBI, ChEMBL, ChemSpider, ConceptWiki,

DisGeNET, DrugBank, Gene Ontology,

neXtProt, UniProt and WikiPathways.

• For questions in drug discovery, answers from

publications in peer reviewed scientific journals.

66

OPS: scientific pharma questions

67

Top questions in the pharma

industry I. (Open PHACTS)

68

Top questions II.

69

Open PHACTS: databases

70

Dataset Downloaded Version Licence Triples

Bio Assay Ontology CC-By 10,360

CALOHA 8 Apr 2015 2014-01-22 CC-By-ND 14,552

ChEBI 4 Mar 2015 125 CC-By-SA 1,012,056

ChEMBL 18 Feb 2015 20.0 CC-By-SA 445,732,880

ConceptWiki 12 Dec 2013 CC-By-SA 4,331,760

DisGeNET 31 Mar 2015 2.1.0 ODbL 15,011,136

Disease Ontology 2015-05-21 CC-By 188,062

DrugBank 19 Feb 2015 4.1 Non-commercial 4,028,767

ENZYME 2015_11 CC-By-ND 61,467

FDA Adverse Events 9 Jul 2012 CC0 13,557,070

Total: ~3 Billion triples

Dataset Downloaded Version Licence Triples

Gene Ontology 4 Mar 2015 CC-By 1,366,494

Gene Ontology Annotations 17 Feb 2015 CC-By 879,448,347

NCATS OPDDR Nov 2015 Oct 2015 2,643

neXTProt (NP) 1 Feb 2014 1.0 CC-By-ND 215,006,108

OPS Chemical Registry 4 Nov 2014 CC-By-SA 241,986,722

HMDB 3.6 HMDB

MeSH 2015 MeSH

PDB Ligands 2 PDB

OPS Metadata CC-By-SA 2,053

UniProt 2015_11 CC-By-ND 1,131,186,434

WikiPathways 20151118 CC-By 11,781,627

Total: ~3 Billion triples

OPS: open tools for free

academic use

73

Open Targets I.

74

https://www.opentargets.org/

Khaladkar, M., Koscielny, G., Hasan, S., Agarwal, P., Dunham, I., Rajpal, D. and Sanseau, P., 2017.

Uncovering novel repositioning opportunities using the Open Targets platform. Drug discovery today.

Koscielny, G., An, P., Carvalho-Silva, D., Cham, J.A., Fumis, L., Gasparyan, R., Hasan, S., Karamanis, N.,

Maguire, M., Papa, E. and Pierleoni, A., 2016. Open Targets: a platform for therapeutic target identification

and validation. Nucleic acids research, 45(D1), pp.D985-D994.

Open Targets II.

75

Linked Data

for the Life Sciences

76

http://bio2rdf.org/

Databases:..

1. Belleau, F., Nolin, M.A., Tourigny,

N., Rigault, P. and Morissette, J.,

2008. Bio2RDF: towards a

mashup to build bioinformatics

knowledge systems. Journal of

biomedical informatics, 41(5),

pp.706-716.

2. Dumontier, M., Callahan, A., Cruz-

Toledo, J., Ansell, P., Emonet, V.,

Belleau, F. and Droit, A., 2014,

October. Bio2RDF release 3: a

larger connected network of

linked data for the life sciences.

In Proceedings of the 2014

International Conference on

Posters & Demonstrations Track-

Volume 1272 (pp. 401-404). CEUR-

WS. org.

Chem2Bio2RDF I.

77

Chem2Bio2RDF II.

78

Artificial intelligence

and

machine learning

in

drug discovery

Drug-target interaction prediction

80

Kövesdi, I., Dominguez‐Rodriguez, M.F., Ôrfi, L., Náray‐Szabó, G., Varró, A.,

Papp, J.G. and Mátyus, P., 1999. Application of neural networks in structure–

activity relationships. Medicinal research reviews, 19(3), pp.249-269.

Colwell, L.J., 2018. Statistical and machine learning approaches to predicting

protein–ligand interactions. Current opinion in structural biology, 49, pp.123-128.

Machine learning

in

chemoinformatics

81

Lo, Y.C., Rensi, S.E., Torng,

W. and Altman, R.B., 2018.

Machine learning in

chemoinformatics and drug

discovery. Drug discovery

today.

Deep learning

in

chemoinformatics

82

Chen, H., Engkvist, O.,

Wang, Y., Olivecrona, M. and

Blaschke, T., 2018. The rise

of deep learning in drug

discovery. Drug discovery

today, 23(6), pp.1241-1250.

Machine learning in chemoinformatics

83

Lo, Y.C., Rensi, S.E., Torng, W. and Altman, R.B., 2018. Machine learning in

chemoinformatics and drug discovery. Drug discovery today.

Results from machine learning

84

Zhang, L., Tan, J., Han, D. and Zhu, H., 2017. From machine learning to deep

learning: progress in machine intelligence for rational drug discovery. Drug

discovery today, 22(11), pp.1680-1685.

De novo molecular design I.

85

Olivecrona, M., Blaschke, T., Engkvist, O. and Chen, H., 2017. Molecular de-novo

design through deep reinforcement learning. Journal of cheminformatics, 9(1),

p.48.

De novo molecular design II.

86

Blaschke, T., Olivecrona, M., Engkvist, O., Bajorath, J. and Chen, H., 2018.

Application of generative autoencoder in de novo molecular design. Molecular

informatics, 37(1-2), p.1700123.

Autoencoder

Variational autoencoder

De novo molecular design III.

87

Blaschke, T., Olivecrona, M., Engkvist, O., Bajorath, J. and Chen, H., 2018.

Application of generative autoencoder in de novo molecular design. Molecular

informatics, 37(1-2), p.1700123.

Generative adversarial autoencoder neural network

Privacy-preserving data analysis

88

Hie, B., Cho, H. and Berger, B., 2018. Realizing private and practical

pharmacological collaboration. Science, 362(6412), pp.347-350.

Chemical syntheses

by deep artificial intelligence I.

89

Segler, M.H., Preuss, M. and Waller, M.P., 2018. Planning chemical syntheses

with deep neural networks and symbolic AI. Nature, 555(7698), p.604.

Chemical syntheses

by deep artificial intelligence II.

90Segler, M.H., Preuss, M. and Waller, M.P., 2018. Planning chemical syntheses

with deep neural networks and symbolic AI. Nature, 555(7698), p.604.

Data and knowledge fusion

in repositioning

Attrition in drug discovery

De novo drug discovery and development

10-17 years process and around 1B USD

~10% probability of success from Phase 1 to Market

Drug repositioning

3-12 years process and up to 80% cost reduction

Significantly higher probability of success from Phase 1 to

Market due to reduced safety and pharmacokinetic uncertainty

De novo discovery vs. repositioning

Scientific motivations for repositioning/rescue

L.A.Barabási:PNAS, 2007,

M.Campillos:Science, 2008Ingenuity Pathway Analysis

A disease-disease similarity network

A drug-drug network

A gene regulatory network

1, Multiple targets

2, Multifactorial diseases

4, Complex pathways (accumulating knowledge)

3, Personalized aspects:

3a, pharmaceutical/phenotypic: efficacy, side effects

3b, genetic/epigenetic

5, New measurements (accumulating omic data)

ENCODE:

tissue specific

regulation

6, Drugome (2000-7000, 1941) + failed drugs (~2000, +100 new yearly)

Scientific motivations for repositioning II.

• Magic bullet vs. Promiscuous/dirty drugs

• Monogenic vs multifactorial disease

• Selective optimisation of side activities (SOSA)

• Network pharmacy

• Personalized („precision”) drugs (for sub-

populations)

– Special external applicability conditions

– „Pathway” drugs

95

Repositioning publications

96

Ashburn TT, Thor KB: Drug repositioning: identifying and developing new uses for existing drugs. Nat Rev Drug Discov

2004, 3(8):673-683.

Campillos M, Kuhn M, Gavin AC, Jensen LJ, Bork P: Drug target identification using side-effect similarity. Science 2008,

321(5886):263-266.

Joachim von Eichborn Manuela S. Murgueitio, Mathias Dunkel, Soeren Koerner,Philip E. Bourne, Robert Preissner:

PROMISCUOUS: a database for network-based drug-repositioning, Nucleic Acids Research, 2010, 1–7

Michael Kuhn, Monica Campillos, Ivica Letunic, Lars Juhl Jensen, Peer Bork,*SIDER: A side effect resource to capture

phenotypic effects of drugs, Molecular Systems Biology 6:343, 2010

……..

0

20

40

60

80

2004200520062007200820092010201120122013

Repositioning: examples

97Li and Jones: Drug repositioning for personalized medicine, Genome Medicine 2012, 4:27

Information sources in repositioning and lead discovery.

Profile Repositioning HTS-based Dimension

Chemical X X 100-10000

Target protein X X n x 10000

Taxonomy X 3 (depth)

Side effect X 10000

Literature X 100000

Gene Expression X X k x 1000

Off-label use X 10000

Chemical fingerprints

• MACCS 2D, Molcon-Z, Dragon, 3D,..

• Schrödinger Canvas using Tanimoto distance

•Structurefingerprint

810 drugs

011001001011010101..

.

001010000001110100..

.

The drug landscape

100

Campillos, M., Kuhn, M., Gavin, A.C.,

Jensen, L.J. and Bork, P., 2008. Drug

target identification using side-effect

similarity. Science, 321(5886),

pp.263-266.

A drug network based on

likelihoods to share targets. All

drug pairs

predicted to share a target with at

least 25% probability were combined

to construct a

drug network. In contrast to Fig. 2 of

the main text, no side-effect similarity

P value

cut-off is used here. The large

network contains 628 nodes (drugs)

with at least 7 side

effects connected by 2881 edges

(drug pairs). Edge width is

proportional to the

probability of the drug pair to share a

target.

The target landscape

101http://www.cytoscape.org/what_is_cytoscape.html

Drug-target landscape

102Keiser, M.J., Setola, V., Irwin, J.J., Laggner, C., Abbas, A., Hufeisen, S.J., Jensen, N.H., Kuijer, M.B.,

Matos, R.C., Tran, T.B. and Whaley, R., 2009. Predicting new molecular targets for known drugs.

Nature, 462(7270), p.175.

Side-effect profiles• DailyMed textmining

– qualitative:SIDER adatbázis (http://sideeffects.embl.de)

– quantitative: exact prevalences

• E.g. Olanzapine

514 drugs

Taxonomies

• Anatomical Therapeutic Chemical

Classification System (ATC)

– 5 levels:

• Main anatomic,

• Main therapeutic

• therapeutic/pharmacological subgroup

• chemical/therapeutic/pharmacological subgroup

• Drugs.com

– http://www.drugs.com/

• RxNorm, Aetionomy 104

Chemical

Target

Pathway

“Disease”

Side effect

drugi

drugj

Combination of chemical and side effect

information for better target prediction

M.Campillos: Drug target identification using side-effect similarity, Science, 2008

Potential avenues of drug repositioning

106Li and Jones: Drug repositioning for personalized medicine, Genome Medicine 2012, 4:27

In silico/virtual screening using LOD

Chemical Side-effects Target prot. MMoA Pathways

Ta

nim

oto

Linked Open Data (LOD), e.g. Open PHACTS

Representation

Surrogate

Compound representations

Compound-compound similaritiesDavis,Shrobe,Szolovits, 1993

Similarity-based virtual screening

1, The “One-One-One” phaseHenrickson J, Johnson M, Maggiori G: Concepts and applications of molecular similarity. 1991, New York: John

Willey & Sons.

Willett P, Barnard J, Downs G: Chemical similarity searching. Journal of Chemical Information and Computer

Sciences 1998, 38(6):983-996.

2, The „data fusion” phase “One-Many-One”Ginn C, Willett P, Bradshaw J: Combination of molecular similarity measures using data fusion. Perspectives in

Drug Discovery and Design 2000, 20(1):1-16.

3, The „group fusion” phase “Many-Many-One”Whittle M, Gillet V, Willett P, Loesel J: Analysis of data fusion methods in virtual screening: Similarity and group

fusion. Journal of Chemical Information and Modeling 2006, 46(6):2206-2219.

Keiser M, Roth B, Armbruster B, Ernsberger P, Irwin J, Shoichet B: Relating protein pharmacology by ligand

chemistry. Nature Biotechnology 2007, 25(2):197-206.

Chen B, Mueller C, Willett P: Combination rules for group fusion in similarity-based virtual screening. Molecular

Informatics 2010, 29(6-7):533-541.

Gardiner E, Holliday J, O'Dowd C, Willett P: Effectiveness of 2D fingerprints for scaffold hopping. Future Medicinal

Chemistry 2011, 3(4):405-414.

Svensson F, Karlén A, Sköld C: Virtual screening data fusion using both structure- and ligand-based methods.

Journal of Chemical Information and Modeling 2011, 52(1):225-232.

B

A

S1S2S3S4 S5

B

A

S*

B

Q2

S S S

B

S*

Q3

Q1

Q2

Q3

Q1

B

S*

Q2

Q3

Q1

B

Q2

S1 S2S3S4 S5

Q1

S2S3S4 S5

Q3

S2S3S4 S5

Q2

Q1

Q3

S*

B

A

S

1, Similarity-based

approach

2, Data fusion

3, Group fusion

4, Query Driven Fusion Framework

Similarity-based fusion in drug repositioning

Chemical Side-effects Target prot. MMoA Pathways

Query-based optimal fusion

Ta

nim

oto

Query: set of corresponding drugs

QDF2

On the use of query analysis

• The information content of

– the query,

– the information resources,

– and the unknown observations(!)

• allow a one-class analysis of the query

(data description)

• and this induction is used in prioritization.

JOINTLY OPTIMIZED:1. weighting the members in the query (e.g. detection of outliers in the question),

GETTING THE RIGHT/IMPROVED QUESTION

2. weighting the similarity measures (e.g.information resources),

GETTING THE SCORING (SIMILARITY) FOR THE RIGHT/IMPROVED QUESTION

3. scoring/ranking the aggregate similarity of the unknown data points to the.

QDF2

The repositome (2013)

The „repositome” of FDA approved drugs (row) for the ATC level 4 classes (columns).

Arany, A., Bolgár, B., Balogh, B., Antal, P., & Mátyus, P. (2013). Multi-aspect

candidates for repositioning: data fusion methods using heterogeneous information

sources. Current medicinal chemistry, 20(1), 95-107.

Drug-target interaction

prediction

Drug-target interaction prediction I.

• Drug/compound information

– Fingerprints, pharmacophore properties, etc.

– Similarities

• Target information

– Protein vs. binding site/pocket

– Sequence/../complete structure

– Similarities

• Interaction data

– Indirect/direct

– Binary/rank/scalar

– IC50, Ki,..

– Complete/incomplete114

Drug-target interaction prediction II.

• Goal

– New drugs for a given target

– New targets for a given compound

– Multitask learning

• Targets for a novel drug

• Drugs for a novel target

• Interaction between novel drugs and targets.

• (Sequentiality)

115

A benchmark DTI task

116

Yamanishi Y, Araki M, Gutteridge A, Honda W, Kanehisa M. Prediction of drug-

target interaction networks from the integration of chemical and genomic spaces.

Bioinformatics. 2008; 24(13):232–40. doi:10.1093/bioinformatics/btn162.

Multitask DTI prediction

• Approaches

– Network methods

– ….

– Pairwise conditional approaches or

pairwise kernel methods

– Matrix factorization methods

117

Fusion of drugs, targets and

interactions

118

Bolgár, Bence, and Péter Antal. "VB-MK-LMF: fusion of drugs, targets and

interactions using variational Bayesian multiple kernel logistic matrix

factorization." BMC Bioinformatics 18.1 (2017): 440.

VB-MK-LMF: performance

119

Bolgár, Bence, and Péter Antal. "VB-MK-LMF: fusion of drugs, targets and

interactions using variational Bayesian multiple kernel logistic matrix

factorization." BMC Bioinformatics 18.1 (2017): 440.

Latent dimensions

120

Bolgár, Bence, and Péter Antal. "VB-MK-LMF: fusion of drugs, targets and

interactions using variational Bayesian multiple kernel logistic matrix

factorization." BMC Bioinformatics 18.1 (2017): 440.

Effect of side information

121

Bolgár, Bence, and Péter Antal. "VB-MK-LMF: fusion of drugs, targets and

interactions using variational Bayesian multiple kernel logistic matrix

factorization." BMC Bioinformatics 18.1 (2017): 440.

Effect of prior

122

Bolgár, Bence, and Péter Antal. "VB-MK-LMF: fusion of drugs, targets and

interactions using variational Bayesian multiple kernel logistic matrix

factorization." BMC Bioinformatics 18.1 (2017): 440.

Unified pharmacogenomic space

123

Bolgár, Bence, and Péter Antal. "VB-MK-LMF: fusion of drugs, targets and

interactions using variational Bayesian multiple kernel logistic matrix

factorization." BMC Bioinformatics 18.1 (2017): 440.

Probabilistic prediction of interactions

124

Bolgár, Bence, and Péter Antal. "VB-MK-LMF: fusion of drugs, targets and

interactions using variational Bayesian multiple kernel logistic matrix

factorization." BMC Bioinformatics 18.1 (2017): 440.

Thank you for your attention!