Regulatory Science Virtual Symposium: “Emerging Technologies in the Medical Device Industry” Session 5: Use of AI in Drug Development (2022)

Course Syllabus/Topics

  1. Speaker background: Medical device lawyer. Helping to figure out how products are regulated, etc.
  2. Typical Drug Development Process:
  3. Where can AI Fit into Drug Development?
    1. Pre-Market:
      • i.      Two main ways used for early drug discovery: use of machines to predict the binding pockets in molecules. Antibody and Protein engineering: very difficult. More so than small molecule development. AI helps more efficiently produce the protein.
      • ii.      Protein engineering: use cases (helps speed up drug development):
      • iii.      Target Discovery
        1. AI helps with screening or compounds. Where in the body can a protein attach? Finding the target. Tumor? Protein? Problematic cell or molecule?
      • iv.      Protein structure
        1. Protein folding/ how does it absorb/distributed:
        2. Knowing the amino acid sequence is a challenge.
        3. The human genome holds 20,000 proteins. We only know part of the structure of these proteins. Have a protein that might turn into a drug. Can use AI to predict the sequence and properties. Identify with AI the optimal experiments to lead to the answer needed.
      • v.      Protein Optimization
      • vi.      Current Examples:
        1. AI tool (DeepMind)
          1. Protein folding is so difficult. Typically, scientists would try to identify the structures, etc.
          2. Predicted the structures of many proteins.
          3. Partnered with a European institute and predicted all cataloged proteins known to science.
        2. Graphics: showing what we knew originally vs what alpha fold identified a year ago, and what they learned in July.
        3. Drug companies are partnering with AI tech.
          1. Healx partnering with Ono: Ono wants to develop drugs for rare diseases.
          2. Models are trained to discover novel disease biology and modes of action.
          3. Biotech Absci used AI models of protein model function manufacturability.
    2. Clinical Trials:
    3. Patient identification:
      • i.      Could identify patients in need of therapy. Some were reviewed by FDA. AI aids in a diagnosis. Marketed therapy for that patient.
  4. Policy issues this raises
    1. Privacy is a key issue to having access to data. EU approach. Big AI regulation.
    2. Key issues on the US side: regulatory and privacy go hand in hand. State laws pop up to address privacy…clinical trials in multiple states have to check if they are following the regulations.
    3. GDPR:
      • i.      Privacy law in Europe. Challenge for any uses of AI and advanced computing. The regulation applies to anyone conducting research.
  1. There is a number of potential targets for new drugs.
  2. How long does it take to bring a new FDA-approved medicine to patients? At least 10 years on average. Oncology drug development is getting faster and faster, but 10 years is the average.
  3. Costs 2.6 billion, on average. Several drugs will get cut in any part of the process including ones that went through the FDA-process
  4. Phases one and two are generally less than phase three. Some drugs for cardiovascular purposes involve following patients that have a cardiac event. No specific breakdown currently (~100 million for one and two)
  1. AI can be used in clinical trials to comb through health records, rare biomarkers, and health history, predicting optimal outcomes.
  2. Certain drugs can help people in other avenues, and these trials sometimes help with that.
  3. 80% of clinical trials fail.
    1. Hard to get enough patients in areas of rare diseases.
    2. In the past clinical trials would request people through referrals and recruitment ads.
    3. Now, we have AI’s ability to extract information.
      • i.      Helpful for patients who may not have had a therapy to treat it.
      • ii.      Patients may be eligible for clinical trials. Identify investigators and site, and compliant with GCP, any violations you have to document, etc.
      • iii.      AI can help look at how past trials have performed.


Accompanying text created by Roxy Terteryan RKS Project Administrator, SC CTSI

NIH Funding Acknowledgment: Important - All publications resulting from the utilization of SC CTSI resources are required to credit the SC CTSI grant by including the NIH funding acknowledgment and must comply with the NIH Public Access Policy.