Harnessing the Power of AI Technology for Drug Discovery Treatments

LuckyStep_AdobeStock

The buzz about the power of artificial intelligence (AI) to either save or doom humanity is rather deafening. A way to explain what people mean by AI is that it comprises the technologies we employ to have computers comb through large datasets using programs (“algorithms”) to find and extract patterns, help humans understand those patterns, suggest decisions regarding future choices, and predict future events or suggest outcomes not previously conceived.

Although some expect AI to mimic human intelligence, some AI systems show performance superior to that of humans (eg, AlphaGo¹). On the other hand, unlike humans, most AI systems need considerable data and training to achieve acceptable performance (eg, we only need to taste strawberries once to know if we like them or not).

One area of research that has attracted a lot of interest and funding is the AI-based design of novel drugs—small molecules, in particular.² A simple way to classify the work being done in this area is shown in the Figure.

Figure. Different Methods for De Novo Drug Design

Focusing on a Target

Target-based approaches rely on an a priori selection of a receptor or enzyme to be activated, inhibited, or modulated.³ Targets for psychiatry, for example, could be monoamine transporters (eg, SSRIs) or newer targets such as TAAR1 (eg, Ulotaront⁴). A receptor model implies knowing or hypothesizing its structure based on known crystalline structures.⁵ A ligand-based model requires a dataset containing the structure of target-selective molecules (“ligands” and “chemical probes”) and the result of their interaction with a target protein (eg, a neurotransmitter receptor) in an appropriate assay (eg, the serotonin 1A receptor⁶).

Importantly, both receptor and ligand-based approaches to de novo drug design implicitly assume that the action at this particular target is the most important action driving the therapeutic value of a drug.

The advantage ofthese modeling platforms is that existing data can be leveraged in silico to predict a new molecule’s action at a target, as well as its physical properties, before triggering costly preclinical work. As the establishment of AI expertise also comes at a considerable cost,various partnerships between AI companies and pharma have been initiated to drive development of AI-based models for de novo drug design, avoiding the need to build up internal AI expertise.

Partnerships such as Genentech/GNS Healthcare, GSK/Insilico Medicine, Takeda/Numerate AI, Atomwise/Abbvie, CrystalGenomics/Standigm, and Cloud Pharmaceuticals/TheraMetrics use novel machine learning techniques to design or identify molecules that act on biological targets of particular interest.⁷ Most of these partnerships, however, do not focus on psychiatric indications or on CNS disorders in general.

Phenotypic Drug Discovery: A Holistic Approach

Drug discovery for mental health, whether traditional or AI-based, has been lagging, whereas mental health needs have continued to increase and remain a great societal burden. Although single target-based AI drug design may be adequate for some indications, the evidence from psychiatry research points to the need of polypharmacology (Figure). Indeed, action at different targets, aiming at an appropriate equilibrium, may be needed.⁸

However, to quantify compound effects at multiple targets in a manner that encompasses downstream actions and interactions, it is necessary to explore drug activity using an in vivo phenotypic drug discovery (iPDD) approach.⁹ An iPDD platform can be used for polypharmacology against known multi targets or in a target-agnostic manner (Figure), as the organism used for drug screening acts as an amplifier of all compound’s actions, providing a comprehensive drug profile. iPDD platforms include those high-throughput screens based on drosophila, zebrafish, and mice.^10-12

Phenotypic screening using an iPDD platform enables characterization of the complete gamut of behavioral effects of reference drugs and a data-driven, target-agnostic comparison with novel compounds. The potential of machine learning-based analysis of behavioral phenotyping data can be leveraged in many drug discovery applications, such as the mining of iPDD screened compounds libraries looking for new hits, or analysis of novel drug candidate analogs, to speed up the lengthy drug discovery process. Furthermore, using animal models of disease in iPDD platforms opens opportunities to explore phenotypic drug screening for psychiatric, neurodegenerative, and rare disorders.¹³

iPDD drug discovery projects can progress in an agnostic manner or turn into a multi- or single-target programs as needed. Although not knowing the mechanism of action makes the road to the clinic harder, target-agnostic programs can be de-risked using “anti-target” panels, such as avoiding D2 antagonism in the development of novel antipsychotics. Biomarkers can also be used to assess target engagement in the clinic when the mechanism of action is unknown.

Perhaps the most convincing evidence will be at hand very soon. Ulotaront, an antipsychotic with a novel mechanism of action now being tested in Phase III, was discovered and developed in partnership between Sunovion and PsychoGenics,⁴ using an iPDD platform (SmartCube®).

Concluding Thoughts

In summary, target- and ligand-based AI de novo drug design approaches have promise for indications with validated hypotheses regarding needed therapeutic mechanisms of action. For complex CNS disorders, on the other hand, phenotypic screens and associated AI methods (such as iPDD platforms) are needed. The phenotypic approach in the upcoming era of AI-based drug design promises to bring new insights and accelerate drug discovery for CNS disorders.

Dr Brunner is chief innovation officer at PsychoGenics Inc. and adjunct associate professor at Mt. Sinai School of Medicine. She is a member of the Business Advisory Board of CTF and the Scientific Advisory Board of CureVCP.

References

1. AlphaGo. DeepMind. Accessed October 12, 2022. https://www.deepmind.com/research/highlighted-research/alphago

2. Vedantam K. AI is making its way into drug discovery. What does it mean for biotech? Crunchbase. October 4, 2022. Accessed October 13, 2022. https://news.crunchbase.com/health-wellness-biotech/artificial-intelligence-venture-drug-discovery/

3. Mouchlis VD, Afantitis A, Serra A, et al. Advances in de novo drug design: from conventional to machine learning methods. Int J Mol Sci. 2021;22(4):1676.

4. Correll CU, Koblan KS, Hopkins SC, et al. Safety and effectiveness of ulotaront (SEP-363856) in schizophrenia: results of a 6-month, open-label extension study. NPJ Schizophr. 2021;7(1):63.

5. Schwartz TW, Hubbell WL. Structural biology: a moving story of receptors. Nature. 2008;455(7212):473-474.

6. Czub N, Pacławski A, Szlęk J, Mendyk A. Do AutoML-based QSAR models fulfill OECD principles for regulatory assessment? A 5-HT1A receptor case. Pharmaceutics. 2022;14(7):1415.

7. Buvailo A. How big pharma adopts AI to boost drug discovery. BioPharmaTrend.com. October 8, 2018. Accessed October 12, 2022. https://www.biopharmatrend.com/post/34-biopharmas-hunt-for-artificial-intelligence-who-does-what/

8. Kondej M, Stępnicki P, Kaczor AA. Multi-target approach for drug discovery against schizophrenia. Int J Mol Sci. 2018;19(10):3105.

9. Leahy E, Brunner D. We need a new Prozac: the demand for brain drug innovation. The Pharma Letter. August 15, 2022. Accessed October 12, 2022. https://www.thepharmaletter.com/article/we-need-a-new-prozac-the-demand-for-brain-drug-innovation

10. Su TT. Drug screening in drosophila; why, when, and when not? Wiley Interdiscip Rev Dev Biol. 2019;8(6):e346.

11. McCarroll MN, Gendelev L, Keiser MJ, Kokel D. Leveraging large-scale behavioral profiling in zebrafish to explore neuroactive polypharmacology. ACS Chem Biol. 2016;11(4):842-849.

12. Roberds SL, Filippov I, Alexandrov V, et al. Rapid, computer vision-enabled murine screening system identifies neuropharmacological potential of two new mechanisms. Front Neurosci. 2011;5:103.

13. Kabitzke P, Morales D, He D, et al. Mouse model systems of autism spectrum disorder: replicability and informatics signature. Genes Brain Behav. 2020;19(7):e12676.