speeding drug discovery with computational … drug discovery with computational biology network...
TRANSCRIPT
Fourteenth International Kidney Cancer SymposiumMiami, Florida, USA—November 6-7, 2015
www.kidneycancersymposium.com
Speeding Drug Discovery with Computational BiologyNetwork Graph Theory
Judge Schonfeld-CEO-CureHunter Inc.
Alexander Schonfeld CTO - Justin Schonfeld CSO - [email protected]/www.curehunter.com
Fourteenth International Kidney Cancer SymposiumMiami, Florida, USA—November 6-7, 2015
www.kidneycancersymposium.com
Computational Biology is easy? About 7-8 Ph.Ds will do nicely.
• Bioinformatics
• Artificial Intelligence
• NLP: Natural Language Processing
• Discrete Mathematics
• Computer Science
• Pharmacology: Clinical Chemistry
• Cell and Molecular Biology
• Genetics
…and it wouldn’t hurt if you were also an MD who actually touched patients.
Fourteenth International Kidney Cancer SymposiumMiami, Florida, USA—November 6-7, 2015
www.kidneycancersymposium.com
Fortunately unifying concepts/partners and tools exist
• Systems Biology Platforms
• Large Centralized Databases
• Off the shelf Math and Graph
• Open Source Data Mining
• Scripting Languages
• Commercial and NPO Partners
• NIH-NSF-DARPA-FDA Fast Track
...it’s all about the need for speed
A conceptual diagram of integrating systems biology platforms using computational tools fornetwork-based drug discovery.
Elaine L. Leung et al. Brief Bioinform 2013;14:491-505
© The Author 2012. Published by Oxford University Press.
Fourteenth International Kidney Cancer SymposiumMiami, Florida, USA—November 6-7, 2015
www.kidneycancersymposium.com
Fortunately Unifying MATH exists: 1 Clinical outcome = 1 Quantum of Knowledge
And just where can I get unified Quanta of Drug, Bioagent, Disease, Biomarkers, Companion Diagnostics and...Clinical Outcomes?
Fourteenth International Kidney Cancer SymposiumMiami, Florida, USA—November 6-7, 2015
www.kidneycancersymposium.com
Read entire National Library of Medicine archive - auto extract evidence of successfuloutcomes: parse - filter - quantify and array data - compute centricity factors
OK…but I can’t read 23 million articles!
Cancer
1…n
Fourteenth International Kidney Cancer SymposiumMiami, Florida, USA—November 6-7, 2015
www.kidneycancersymposium.com
But Computational Linguistics* algorithms CAN read them all--in about 4 hours--and automatically extract outcome-based drug-to-disease MOA results:
“Cetuximab is a chimeric immunoglobulin G1 monoclonal antibody that
targets the extracellular domain of the epidermal growth factor receptor
(EGFR) with high specificity and affinity to achieve high clinical efficacy
in several types of cancers, particularly colorectal and head and neck cancer.”
• How Much do you know from reading that 1 sentence?
• How Much do you know if you read 230,000,000 of them and extract 10-20 key variables from each statement?
• If you “connected” all the dots could you or your AI Machine discover a new cure for Breast Cancer or Ebola in 4 minutes? YES…you could…we did.
*Conceived originally to decode the Nazi Enigma Machine such algorithms were highly developed over the next 70 years for machine translation, speech processing, DNA sequencing and many other applications.
Fourteenth International Kidney Cancer SymposiumMiami, Florida, USA—November 6-7, 2015
www.kidneycancersymposium.com
What if a computer could tell you in 10 seconds every drug that ever achieveda significant remission in any and all peer-reviewed cancers? [The CureHunter Machine can]
What pattern might you see? What protocols might be optimized?
“We report a case of poor-risk metastatic renal cell carcinoma, with Von Hippel-Lindau loss of function, which achieved and maintained a complete remission after first-line therapy with sunitinib by using a reduced dosage and a modified schedule of treatment.” Anti-cancer drugs (Anticancer
Drugs) Vol. 26 Issue 4 Pg. 469-73 (Apr 2015) ISSN: 1473-5741 [Electronic] England
Efficacy evolves toward certainty over time
upward trending knowledge indicates
compound potential…often years in
advance of a formal trial
Fourteenth International Kidney Cancer SymposiumMiami, Florida, USA—November 6-7, 2015
www.kidneycancersymposium.com
In human discovery we call this connecting the dots…and what’s good for Intelligent Man is good for Intelligent Machine
In the head of the Virtual PI
247,600 drugs and bioagents X
11,600 disease states X
15,000,000 outcomes for1.2 billion patients
are networkedthen
vectored towardnew cures
Neuron or Network Graph Engine
Fourteenth International Kidney Cancer SymposiumMiami, Florida, USA—November 6-7, 2015
www.kidneycancersymposium.com
So just What does an AI
Computer
see in this mess, exactly?
The same things any good
scientist does:
Patterns, cross connects,
new hypotheses testable
by computations for:
association
correlation and causation
Fourteenth International Kidney Cancer SymposiumMiami, Florida, USA—November 6-7, 2015
www.kidneycancersymposium.com
So if I connect the dots: Can a Breast Cancer Drug Stop Ebola?
"From this screen [*HTS well plate], we identified a set of selective estrogen receptor modulators (SERMs), including clomiphene and toremifene, which act as potent inhibitors of EBOV infection.”Sci Transl Med. 2013 Jun 19;5(190):190ra79. doi:
10.1126/scitranslmed.3005471
The CureHunter HTKS* Computational Biology Engine
screened 230,000,000 sentences in its silicon model of the
entire Medline Archive to predict the same result in 4 minutes
of computation.
*CureHunter HTKS: High Throughput Knowledge Screening is an in silico analog to HTS wet chemistry well plate analyzers that are used for comparators and validation.
Fourteenth International Kidney Cancer SymposiumMiami, Florida, USA—November 6-7, 2015
www.kidneycancersymposium.com
CureHunter solutions focus on Graph Isomorphism between Drug-Disease-Biomarker Outcome Networks: Think Fingerprint/Facial Recognition matching
To date the CureHunter Engine has successfully predicted and validated 719 new cures for human disease with Network Graph Theory
Fourteenth International Kidney Cancer SymposiumMiami, Florida, USA—November 6-7, 2015
www.kidneycancersymposium.com
Is “Big Data” then just market hype? No. But the devil…as usual…is in the scientific details.
BIG data is not always GOOD dataWhat’s wrong with using Google, EMRs, Billing Records, Patient self-reportage, Facebook sentiment “Drug Likes” et al in Network Graph Modeling?
• Lack of Controls for precision, accuracy, margin of error, source input QA, population normalization, replication/validation• Lack of Compliance tracking• Lack of Clinical Outcome tracking• Lack of peer Review• Noise = High - Signal = Low• Worst Case = Salted and Apples to Oranges
You want BIG & GOOD IN = GOOD PREDICTIONS OUT
Without these you have
information
but you don’t have
science
Fourteenth International Kidney Cancer SymposiumMiami, Florida, USA—November 6-7, 2015
www.kidneycancersymposium.com
In Sum: Outcome-Centric Data Mining Works by Converting 200+ years of
Clinical Evidence in natural language archives with millions of outcomes for--billions of patients--to Quantified Network Diffusion Graphs Automatically
Chi extracted Positive
Clinical Outcome Data
from US NLM
1809 - To Time Now
Updated daily
• Create Numeric Profiles for all successful med-disease relationships • Compare Profiles and predict future new indication clinical utility• Deliver analysis package to Pharma/University/Med Center partner turnkey
• # of useful meds
• Mechanism of action
• Key active targets
• Key markers-diagnostics
• Role in multiple diseases
• Pathology centricity
• Cure centricity of agents
• Safety-toxicity data
Fourteenth International Kidney Cancer SymposiumMiami, Florida, USA—November 6-7, 2015
www.kidneycancersymposium.com
Who’s doing good graph work? Worldwide about 600 scientists. Below Marc Vidal/Dana Farber & Albert-Laszlo Barabasi/Northeastern
Paradigm article Published in Proceedings of the National Academy of Sciences: The Human Disease Network, PNAS May 22, 2007 vol. 104 no. 21 8685–8690
Fourteenth International Kidney Cancer SymposiumMiami, Florida, USA—November 6-7, 2015
www.kidneycancersymposium.com
Clinical Translation of Genetic Graphs: Bench to Bedside Dr. Justin Schonfeld/CureHunter: Map Gene to Marker to Drug to Outcome
[Schematic Algorithm for Personalized Medicine]
Does patient have GENE?
If YES, does patient have DISEASE?
If YES, does DISEASE have drug?
If YES, is drug EFFECTIVE?
If YES, treat
IF NO, go to DISCOVERY MODE:
Does GENE network to other diseases?
IF YES, do other diseases have drugs?
IF YES, can they be repurposed, prescribed off label?
IF YES, treat
IF NO, does PROTEIN, MARKER or CD network….
IF YES, proceed to discovery level II:
Can protein expression be altered?
IF YES, which molecules/MOA?
Small graph segment only
Fourteenth International Kidney Cancer SymposiumMiami, Florida, USA—November 6-7, 2015
www.kidneycancersymposium.com
Computational Biology can meet the need for speedthrough integrated digital modeling Art-to-part of the Molecule-to-Medicine pathway:
Time to Patient/Concept to Clinic 10X reductions…a new cancer therapy while you wait. How?
Computationally compress the first 6 years of R&D…that’s really Fast Track
0 to Phase I = 4.8 minutes
Thank You: To partner on chemotherapy protocol optimization or drug discovery please contact: [email protected]
Emulate mechanical engineering CAD-ME: Computer Aided Design with CA-DDD: Computer Aided Drug Discovery and Development