Connector Corner: Accelerate revenue generation using UiPath API-centric busi...
BenevolentTech - Harnessing the power of AI to accelerate global scientific discovery
1. We do it because
it matters
Jerome Pesenti, BenevolentTech CEO, BenevolentAI
2. WHAT’S the problem?
Despite the huge growth of knowledge
the process of scientific discovery HAS
NOT CHANGED FOR 50 YEARS
In the modern world it’s IMPOSSIBLE
for humans alone to process all the
complex information available to
ensure continued significant advances
in scientific discovery
Without technology’s help science
based industries will REGRESS
15/03/2017 2
3. ABOUT BenevolentAI
A British technology Company applying AI to
accelerate and enhance scientific discovery
15/03/2017 3
Founded in 2013 – post successful exit of Proximagen Group plc
$BN company with strong balance sheet ($87m raised to date)
70+ team of world class scientists and technologists (40+ doctorates/advanced degrees)
Technology enabling previously impossible tasks in bioscience
Across the road from here
5. HOW we are different
15/03/2017 5
Platform Vendors (e.g. Watson)
• Scientific discovery is one of many use cases
• Focus on bottom of the stack rather than
solving use case end-to-end
• Technology developed in a vacuum without
direct access to scientists
• Ecosystem strategy with horizontal technology
that has limited understanding of domain
knowledge
• No user feedback loop, developed in isolation
• Sell software
BenevolentAI
• Sole focus on a specific discovery use case: end
to end drug discovery and development
• Direct and constant access to the science team
• Laser-targeted to the needs of the drug
discovery scientists
• Understands very precisely domain & language
that can extract and represent existing
knowledge
• Uses domain knowledge to make predictions
• Improves and learns quickly based on user
feedback, building unique differentiation
• Applying software to develop own drugs
6. OUR traction and technology validation
24 validated hypothesis in clinical and pre clinical development in 6 months
$800m signed out-licence agreements
300granted patents (portfolio) covering large series of compounds
MULTIPLE repurposing opportunities being pursued in clinic (J&J in-licence)
1st Phase IIb clinical study in 2017
15/03/2017 6
7. Our AI enabled drug discovery workflow
15/03/2017 7
1. Initial
Research
2. Hypothesis
Generation
3. Hypothesis
Validation
4. Programme
Initialisation
Drugs
Targets
Diseases
Novel associations
Rapid in silico
validation
Biological testing
Private and Confidential
8. KNOW WHAT YOU KNOW – in order to put the
pieces together but first we need to know what
the pieces are!
BREAKING SILOS - most biological knowledge is
locked away in scientific publications/patents/
reports. Each individual scientist only knows a
small subset of this.
SUM OF THE PARTS - by automating the
extraction of biological knowledge, we can enable
our scientists to base their work on all biological
knowledge rather than a small slice and find
connections they would otherwise have missed.
15/03/2017 8
MACHINE LEARNING AND NLP - need to use and
develop state-of-the-art NLP and machine learning
algorithms.
UNIQUE ALGORITHMS - making predictions
requires using unique deep learning algorithms that
leverage information across multiple data sets,
both structured and unstructured.
UNIQUE TOOLS - partnership with drug scientists
requires uniquely designed tools and the efficient
collection and processing of feedback.
BIO-RELEVANT - deep learning algorithms need
to be adapted to bio-chemical structures and
processes.
The BenevolentAI tech challenges
Private and Confidential
9. Our methodology and the skills it requires
15/03/2017 Private and Confidential 9
Codify drug
development
Implement
discovery product
Optimize discovery
steps
Track usage
measure
performance
• Drug Discovery
• UX Design
• Product Mngt
• Software Engr
• ML, NLP, DL
• Bio/Chem
Informatics
• Data Science
Collect feedback
• Machine Learning
• UX Design
10. WE ARE RECRUITING!
• 40 AI researchers in London and NYC.
• 10 software engineers in London.
15/03/2017 10