Recent Successes in Phenotypic Drug Discovery and the Future of ML/AI Methods

Table of Contents:

1. History of Phenotypic Drug Discovery
2. Advantages of Phenotypic Drug Discovery
3. Recently Approved Treatments Identified through Phenotypic Drug Discovery
4. AI Tools for a New Era of Phenotypic Drug Discovery

Phenotypic drug discovery has contributed to the development of dozens of novel treatments in recent years, many being first-in-class medicines. This drug discovery approach uses screening methods that do not require knowledge of the molecular target. Instead, it relies on identifying phenotypic changes associated with the effect of an active molecule on cells, tissues or whole organism systems relevant to human disease. 

Phenotypic drug discovery process involves using a diverse repertoire of methods, such as different types of cellular assays or high-content screening (HCS). The latter is arguably the most powerful and has been further enhanced by the recent advances in computational image analysis, including machine learning (ML) and Artificial Intelligence (AI) approaches. 

In contrast to target-based drug discovery, which relies on the meticulous investigation of a single target, phenotypic drug discovery empowers researchers to screen compound libraries against thousands of potential targets in a single experiment. This unbiased investigation method promotes the discovery of novel mechanisms, targets, pathways and lead molecules. Additionally, testing molecules directly in living systems that mimic disease states presents a significant advantage for generating insights that are more relevant to clinical outcomes [1].

History of Phenotypic Drug Discovery

Renewed interest in phenotypic drug discovery was an outcome of a systematic analysis of the list of new FDA-approved treatments between 1999 and 2008. A review published in Nature Reviews Drug Discovery, reports that phenotypic drug discovery methods were responsible for 28 first-in-class small molecule drugs discovered compared to 17 from the target-based methods [2]. Another report by Haasen et al [3] showed a dramatic increase in the percentage of phenotypic screens done by Novartis from 2011 to 2015. 

From 2012 to 2022, application of phenotypic drug discovery methods has grown from less than 10% to an estimated 25-40% of the project portfolio of large pharma companies such AstraZeneca and Novartis [4]. The success of this pivot has been evident in the track record of recently approved treatments developed through phenotypic drug discovery efforts. 

Advantages of Phenotypic Drug Discovery

An important benefit of phenotypic drug discovery is that its unbiased nature allows for the identification of therapeutic interventions for novel and diverse targets. In contrast to target-based drug discovery, which typically goes after enzymes and receptors, phenotypic approaches can reveal   therapeutic intervention routes acting via membranes, ion channels, ribosomes, microtubules, or large complex molecular structures like the ATP synthase, as well as other unknown targets. 

With such a wide diversity of novel targets and mechanisms, phenotypic drug discovery projects have identified game-changing medicines, including first-in-class treatments for Duchenne muscular dystrophy, spinal muscular atrophy, cystic fibrosis, hepatitis C virus, and others. 

Recently Approved Treatments Identified through Phenotypic Drug Discovery

Vamorolone. One of the most recent examples of medicines that were identified through phenotypic drug discovery is a therapy for Duchenne muscular dystrophy, one of the most severe forms of hereditary neuromuscular diseases. A potent antagonist of the mineralocorticoid receptor (MR), Vamorolone (AGAMREE®), was developed by Santhera Pharmaceuticals and received approval in 2023. This first-in-class drug binds to the same receptors as corticosteroids but modifies the downstream activity of the receptors by ‘dissociating’ efficacy from typical steroid safety concerns, making it a valuable alternative to corticosteroids for children and adolescent. In the case of vamorolone, phenotypic profiling enabled the elucidation of the sub-activities of this drug [5]. 

Risdiplam. Another recently approved medicine that was developed using phenotypic drug discovery methods is Risdiplam, a Spinal Muscular Atrophy (SMA) treatment that can be administered at home. Risdiplam (sold under the brand name Evrysdi®) was developed by Genentech in collaboration with PTC Therapeutics and the SMA Foundation and approved for use in adults, children, and infants by the FDA in 2020. This systemically distributed small-molecule drug works through the modulation of SMN2 pre-mRNA splicing and increases levels of full-length SMN protein.  Due to lack of known activity, SMN2 would have been an unlikely target in a traditional, target-based drug discovery campaign [6]. 

Table 1: Recently approved therapies identified using phenotypic drug discovery methods

Daclatasvir (Daklinza) was developed by Bristol-Myers Squibb and approved in the European Union in 2014 and the USA in 2015. This antiviral agent is used in combination with other medications to treat hepatitis C (HCV). The drug was identified through phenotypic screening and later revealed to target NS5A, a non-structural protein that plays a key role in the HCV replication process [7]. As a protein with no enzymatic activity and its mechanism of action being not fully understood, it remained an elusive target for many years. Daclatasvir became first in the class of NS5A inhibitors to reach the market. A common thread between Daclatasvir and Risdiplam is that both drugs’ targets lack known activity and functional roles in disease and are therefore unlikely to be identified by traditional drug discovery methods. 

Lumacaftor. Another molecule identified through phenotypic drug discovery is lumacaftor, which used in combination with ivacaftor as a therapy for cystic fibrosis and marketed under the name ORKAMBI®. Developed by Vertex Pharmaceuticals, it was approved for medical use in the United States in 2015. It targets the defective transmembrane conductance regulator (CFTR) and is effective at treating the disease in patients homozygous for the F508del mutation in CFTR gene. The molecule was discovered using target-agnostic compound screens that used cell lines expressing wild-type or disease-associated CFTR variants [8]. 

Perampanel, sold under the brand name Fycompa, is an epilepsy treatment that was developed by Eisai Co., Ltd. This drug was first approved in 2012 to treat partial seizures and generalized tonic-clonic seizures for people older than twelve years. Whole-system, multi-parametric modeling was needed in the development of perampanel, which is not a common approach used in traditional target-based drug discovery [9]. 

If we take a look at the list of new treatments approved from 1999 to 20173, phenotypic drug discovery has contributed to the development of 58 out of 171 total drugs. Traditional target-based drug discovery follows the lead, with 44 approvals, and monoclonal antibody (mAb)-based therapies responsible for 29 approvals. In addition, numerous candidates identified through phenotypic drug discovery programs are currently in clinical trials, suggesting the numbers will continue to increase.  

Figure 1: New therapies from different discovery strategies (1999-2017). Data sources: Swinney and Anthony (2011); Haasen, et al. (2017). 

AI Tools for a New Era of Phenotypic Drug Discovery

Over the last two decades, phenotypic drug discovery contributed to a large number of first-in-class medicines with novel molecular mechanisms. It has proven to be a powerful strategy for identifying new molecular entities that are not derivatives of existing or previously investigated compounds. This enabled the discovery of therapies that have provided valuable interventions for unmet medical needs, such as Duchenne muscular dystrophy, spinal muscular atrophy, hepatitis C, and cystic fibrosis. 

To further advance phenotypic drug discovery, the biotech industry is dedicated to improving the methods for phenotypic profiling, developing better disease models and novel computational approaches. In particular, ML and AI tools provide significant advantages for phenotypic drug discovery by enabling automated analysis of cell image data, extraction of diverse morphological features and clustering of cellular phenotypes to help identify potential drug candidates, as well as for many other applications.

In addition, advanced computational methods can take advantage of multimodal data. For example, using chemical structure features and together with extracted image features for elucidating the Mode of Action (MoA) and bioactivity properties significantly improves the prediction power of the method. Diverse types of large public datasets have become a valuable data source for phenotypic drug discovery. The phenotypic drug discovery community is collaborating by sharing datasets and analysis methods through consortia such as JUMP-CP. Ardigen has been a supporting partner of the JUMP-CP, contributing resources such as the JUMP-CP Data Explorer tool. 

Incorporating machine learning and AI solutions into phenotypic drug discovery workflows can provide additional dimensionality and powerful insights, improve the success rate and accelerate the speed of drug discovery, as well as provide access to the greatest diversity of target types and identify novel mechanisms.

Explore Ardigen’s phenAID platform, dedicated to reducing the analysis time and enhancing the quality of predictions for HCS datasets. Visit our website to learn more. 

Works Cited: 

1. Swinney, David C., and Jonathan A. Lee. “Recent advances in phenotypic drug discovery.” F1000Research9 (2020). https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7431967/
2. Swinney, David C., and Jason Anthony. “How were new medicines discovered?.” Nature reviews Drug discovery10.7 (2011): 507-519. https://www.nature.com/articles/nrd3480
3. Haasen, Dorothea, et al. “How phenotypic screening influenced drug discovery: lessons from five years of practice.” Assay and drug development technologies15.6 (2017): 239-246. https://www.liebertpub.com/doi/abs/10.1089/adt.2017.796
4. Vincent, Fabien, et al. “Phenotypic drug discovery: recent successes, lessons learned and new directions.” Nature Reviews Drug Discovery21.12 (2022): 899-914. https://www.nature.com/articles/s41573-022-00472-w
5. Wells, Elizabeth, et al. “Vamorolone, a dissociative steroidal compound, reduces pro-inflammatory cytokine expression in glioma cells and increases activity and survival in a murine model of cortical tumor.” Oncotarget8.6 (2017): 9366. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5354737/
6. Ratni, Hasane, et al. “Discovery of risdiplam, a selective survival of motor neuron-2 (SMN2) gene splicing modifier for the treatment of spinal muscular atrophy (SMA).” (2018): 6501-6517. https://pubs.acs.org/doi/full/10.1021/acs.jmedchem.8b00741
7. Belema, Makonen, and Nicholas A. Meanwell. “Discovery of daclatasvir, a pan-genotypic hepatitis C virus NS5A replication complex inhibitor with potent clinical effect.” (2014): 5057-5071. https://pubs.acs.org/doi/full/10.1021/jm500335h
8. Van Goor, Fredrick, et al. “Correction of the F508del-CFTR protein processing defect in vitro by the investigational drug VX-809.” PNAS108.46 (2011): 18843-18848. https://www.pnas.org/doi/abs/10.1073/pnas.1105787108
9. Hanada, Takahisa. “The discovery and development of perampanel for the treatment of epilepsy.” Expert opinion on drug discovery 9.4 (2014): 449-458. https://www.tandfonline.com/doi/full/10.1517/17460441.2014.891580