Hemodialysis: Chapter 1, Physiological Principles of Hemodialysis - Dr.Gawad
AI_healthcare_Dr_Stefan_Pfeiffer
1. How data science and artificial
intelligence will shape the future of
drug development and approval
Dr. Stefan Pfeiffer, 28.6.2019
2. • machine vs physicians?
• “narrow AI”
• Algorithms to solve specific tasks
• Different approaches depending on data types
• Supervised learning
• Unsupervised learning
Roboter Xiaoyi. Picture: iFlytek
AI in healthcare, what is it about?
Dr. Stefan Pfeiffer, 28.6.2019
3. Applications of AI in drug discovery and
development?
Target discovery
Building predictive models to uncover complex patterns of
biomarkers and resistance mechanisms (e.g. Bayesian
modelling)
Drug design
Improved molecule design in silico (e.g. Schroedinger)
Clinical trials
Predict drug response in patients (faster and better clinical
trials)
Patient treatment
Personalization of treatments based on similarities between
patients (e.g. clinical or demographic metadata)
Dr. Stefan Pfeiffer, 28.6.2019
4. • Need for efficiency and efficacy
1 in 10 small molecule projects become candidates for clinical trials -> 1 in 10
pass clinical trials.
• Need for speeding up and lowering costs
Drug development takes on average about 15 years, costs > $1 billion.
• Need for novel drugs and drugs for so far untreated diseases
Traditional drug development process based on personal knowledge (target
driven), modelling makes use of available information (data driven).
Why there is a need for AI in drug discovery and
development?
Dr. Stefan Pfeiffer, 28.6.2019
5. Benefits
• More efficient study design-> algorithms that provide relevant
information to scientists for study design
• Improved objectivity(e.g. predetermined targets when screening
data).
• Higher predictive power to define meaningful interactions in screening
-> reduction of false positives
• In silico drug screening (pre-screening) -> more novel and more
promising candidates.
• More efficient reducing time-scale and cost of drug discovery
Dr. Stefan Pfeiffer, 28.6.2019
6. Weaknesses, Threads
• AI predictions are as good as the algorithms used to investigate a dataset.
• Algorithm based on defined assumptions that are made by science based on
a smaller dataset -> can lead to grey zones, wrong assumptions.
• Algorithm bias: bias, shortcomings.
• Standardization vs. development
• Investment costs for infrastructure, pipeline development.
• DATA SECURITY & PRIVACY IS A MAJOR CONCERN
• At this point, all predictions have to be verified by a scientist.
Dr. Stefan Pfeiffer, 28.6.2019
7. International cooperatives for standardisation in drug discovery
New players might enter the field
• Cooperative initiatives, e.g. MELLODDY (EU project, academia & big
pharma -> standardized algorithms)
• Non-traditional healthcare companies e.g. Google Brain (Google LCC)
Dr. Stefan Pfeiffer, 28.6.2019
8. • R&D efficiency of biopharma companies is declining
• R&D costs, legal requirements, short drug life span (e.g. antibiotic resistance),
political pressure on drug prices, high rate of commercial failures.
• AI can provide a decrease of costs for R&D, clinical trials.
• AI can help to develop better and more personalized drugs.
• Currently, AI healthcare companies prevalently in US and UK, the EU lags
behind. (~300 AI healthcare start-ups founded in recent years)
• Companies that implemented AI in their drug development R&D had
increased revenue (e.g. Pfizer).
• Expected CAGR AI healthcare: 68%, 2018-2022 (Frost & Sullivan, 9/2018)
• How BE can profitate from AI expertise? Not only drug development..
SWOT including weakness associated with a lack of AI -> advice,
strategy, building ressources.
Outlook
Dr. Stefan Pfeiffer, 28.6.2019