Machine Learning AI in Drug Discovery Medicinal Synthetic

Machine Learning / AI in Drug Discovery Medicinal & Synthetic Chemistry Perspective Dr Ed Griffen Technical Director Med. Chemica

Machine Learning / AI in Drug Discovery Medicinal & Synthetic Chemistry Perspective Organization TI LG LO Scientist Data and Technology Med. Chemica 2 Ed Griffen 2018

ML in Chemistry Maturity • > than any chemist & database • = an experienced chemist with a large Pharma database • = experienced chemist • Random – no effect 3 Med. Chemica • = graduate chemist

ML in Chemistry Maturity Virtual Screening Synthesis route Potency design Optimization LG 4 Ed Griffen 2018 ADME Tox Optimization Alerts LO Polymorph Prediction Med. Chemica $ Value

Data and Technology Open source and cloud driving the revolution Data Machine Learning • • • 5 zero barrier to entry instantly scalable massive user base Ed Griffen 2018 • • DB technologies ubiquitous Public data growing Large volumes of well curated data essential Libraries ubiquitous Large user base Med. Chemica Cloud

Data and Technology Only Big Pharma have enough data? NLP – Natural Language Processing 6 Ed Griffen 2018 Med. Chemica Benevolent. AI

ML in Chemistry Maturity Virtual Screening ADME Optimization Synthesis route design Tox Polymorph Prediction Alerts Potency Optimization 7 Ed Griffen 2018 Med. Chemica $ Value

ADME Optimization Making a real textbook of Medicinal Chemistry MMPA Better Project decisions Combine MMPA and Extract MMPA Astra. Zeneca, Roche and Genentech ADMET data Rules >437000 rules Increased Medicinal Chemistry learning Med. Chemica Learning Medicinal Chemistry Absorption, Distribution, Metabolism, Excretion, and Toxicity (ADMET) Rules from Cross-Company Matched Molecular Pairs Analysis (MMPA). Kramer, Robb, Ting, Zheng, Griffen, et al. J. Med. Chem. 2017 http: //pubs. acs. org/doi/10. 1021/acs. jmedchem. 7 b 00935

Synthesis Route Design Automated Retrosynthetic analysis comes of age Med. Chemica

Scientists Help the Hi. PPOs – or they’ll crush you “Companies often make most of their important decisions by relying on “Hi. PPO”—the highestpaid person’s opinion. ” 1 Chemistry Hi. PPs: 10 experts in pattern recognition judged on their ability to make the best decisions with partial data highly trained time poor delivery focused gatekeepers to the adoption of new approaches Ed Griffen 2018 1. Mc. Afee & Brynjolfsson “Big Data: The Management Revolution”, Harvard Business Review October 2012 Med. Chemica • • •

Scientists Is ML in chemistry mature enough yet? Replace the chemist with automation? Advantages Propose Compounds Speed Med. Chemica Prediction & error prediction Lack of bias Cost per compound Sort &Filter Med. Chemica Lee Cronin Nature (559) , 377 – 381 (2018) Fragile models Unclear where they fail Un-auditable Scalability No explanations needed! 11 Risks

Scientists 12 Ed Griffen 2018 Med. Chemica Christian Tyrchan AZ Gothenberg

Scientists Chemists enhanced by Computational Intelligence Advantages Propose Compounds Med. Chemica Integrating experience context Prediction & error prediction Sort &Filter Critical analysis Stopping ‘stupid’ errors Risks Slower: humans in decision loop Critical: Person – machine interface 13 Lee Cronin Nature (559) , 377 – 381 (2018) Med. Chemica Complexity

Scientists Med. Chemica 14 Ed Griffen 2018

Organization • Machine Learning challenges (powerful) cognitive workers • Technologies cross previous organizational boundaries • Skunkworks • Acquisitions 15 Ed Griffen 2018 Med. Chemica Options to address internal organizational resistance to change:

Machine Learning / AI in Drug Discovery Medicinal & Synthetic Chemistry Perspective Summary • AI is delivering on the back of • massive data sets, • turn-key high performance computing • ubiquitous machine learning libraries • choose: build it or buy it 16 Ed Griffen 2018 Med. Chemica Challenges • Knowledge workers are more challenged by AI • Integration or automation • Organizational change