Dr Lotfi Chouchane and Dr. Javaid Sheikh, Weill Cornell Medicine-Qatar, outline the challenges and opportunities ahead for drug discovery
Precision medicine is an emerging model for the next generation of clinical care that will capitalise on the dynamic interaction between individual biology, lifestyle, behaviour, and environment. It holds huge promise for healthcare and the drug discovery sector in particular.
An essential objective of precision medicine is quantifying an individual’s risk for any disease and tailoring personalised prevention and therapeutic strategies. This includes improving diagnosis, designing therapeutic interventions and determining prognosis through the use of large complex datasets that incorporate individual gene, function and environmental variations1. A targeted and more effective drug design based on the integration of multiple sources of data for each individual, including longitudinal multi-omic datasets, can lead to more personalised treatments.
There is much progress still to be made before the potential of precision medicine is fully realised. The recent Economist Intelligence Unit report on precision medicine2, commissioned by Qatar Foundation, highlighted several challenges for precision medicine implementation, including the challenge of harnessing the vast data pools that already exist in order to produce actionable insights for clinicians within health systems.
Nevertheless, artificial intelligence (AI), machine learning (ML), deep learning (DL) and big data analytics are evolving to be a great aid to precision medicine. Furthermore, information and communication technologies in general and wearable sensors are helping to promote a greater level of precision in healthcare3.
Precision medicine and biobanks
Precision medicine uses diverse technologies to collect and interpret personalised data for the sole purpose of an individual’s treatment. The ability to use intelligent algorithms to mine vast stores of unstructured and structured data for better insights has empowered providers with the tools to design personalised interventions for individual patients. Recent advancements in the fast collection of data has led to an incredible increase in the volume of biological and medical data collected from human populations, with UK Biobank4, the “All of Us” research program5and the China Kadoorie Biobank6 generating extremely thorough and deep phenotypic reports of health trajectories for millions of individuals.
Similarly, personalised healthcare initiatives in Qatar are part of a coordinated and comprehensive precision medicine strategy to deliver world-class future healthcare. The Qatar Biobank has been conducting a large population-based cohort study, which was initiated in 2012 by Qatar Foundation. The biobank’s broad data sets already cover exposomes and whole genome sequencing from 20,000 individuals7, and Qatar Genome Programme plans to sequence complete genomes of around 300,000 native Qataris. This will provide multitudes of rich source data to fulfil the aims of applying and advancing precision medicine powered with AI in Qatar8.
Recently, the Qatar National Research Fund (QNRF), the main research funding agency in Qatar, and Qatar Genome Programme also launched the ‘Path Towards Precision Medicine’ research programme. This initiative aims to support genomics research to promote drug discovery and use patient specific genomic variants for tailored, personalised therapies for the Qatari population.
AI and machine learning in drug discovery and development
Historically, drug discovery is a long and very expensive process and highly prone to failure due to unexpected/unpredictable toxicity, poor pharmacokinetics or insufficient activity of potential therapeutic molecules. The launch of a new drug on the market costs between several billion to tens of billions of dollars, typically taking between three to 20 years. A research survey among 106 new drugs developed by 10 pharmaceutical companies found that they cost on average $2.7 billion (£2 billion) to develop9.
In the current omics era of big data, the implementation of AI/ML based algorithms has leveraged the paradigm of ‘one gene, one target, one drug’ into a framework of unselective targets, even for one drug10. In this context, AI/ML/DL algorithms can learn from heterogeneous datasets and discover new drug targets, repurpose the current existing ones or eventually guide the decision-making protocol. Recently, this was demonstrated with international clinical data sharing programmes for COVID-19, which has opened up an enlightened vision in which AI/ML can guide clinicians to a rapid classification of the severity of the infection and thus the most effective treatment11.
Moreover, using AI/ML can enable the drug discovery community to benefit from large sets of expression data from target tissues or organs. This data can help to identify cell membrane receptors with a regulatory role in disease-related gene expression. This allows medicinal chemists to elucidate the mechanism of action of a disease, trace back the target(s), data mine existing databases for drugs with an inhibitor, and computationally predict the effectiveness, potency and selectivity of different drugs.
For example, in an effort to improve the accuracy of deep DTnet – a deep learning algorithm for the identification of new drug targets – Zeng and co-workers experimentally predicted and validated topotecan, an approved drug for ovarian carcinoma, as a promising treatment for multiple sclerosis12. In addition, a recent study into the inhibitors of triple-negative breast cancer, which made use of deep neural networks (DNN), demonstrated which compounds were most efficient13. This implies that AI/ML has a chance to be applied in drug selection, repurposing and thus accelerating the process of drug discovery without repeatedly reverting to de novo design.
We have also been seeing growing interest of big pharma companies in applying AI/MI in precision medicine for drug discovery and treatment. Aiming to apply genome sequencing powered by AI/ML to a large population, the pharma company Roche acquired private company Bina in 2014. Additionally, GNS healthcare announced the launch of a collaboration with Genetech to boost the development of novel cancer therapies using the GNS REFS (Reverse Engineering and Forward Simulation) causal machine learning and simulation platform. Furthermore, the progressive partnership between Biogen, EMBL-European Bioinformatics Institute, GlaxoSmithKline and the Wellcome Trust Sanger Institute established the Open Targets validation platform.This platform is a public-private partnership that uses genetics and genomics data for systematic drug target identification and prioritisation, and this large database trained on four different ML classifiers. Medicinal chemists can now use the platform for drug discovery14.
AI and machine learning in precision oncology
Precision oncology treatments rely heavily on the patient’s genomic data to make treatment decisions. Whole genome sequencing has already improved our understanding of tumours; the unprecedented molecular detail has enabled highly efficient targeted therapy, coupled with a new generation of drug development.
At the same time, cancer drug development is evolving rapidly thanks to precision medicine, with efforts centred on matching drugs or treatments to predictive marker(s) for selected patients15. The use of AI/ML has already proven to be successful in choosing the drug combination based on a patient’s own biopsy, and to make recommendations for N-of-1 medications 16. In cancer treatment, it is crucial to identify reliable drug targets and the driver genes for personalised medicine. AI/ML has begun to play a role in generating novel drug candidates and repurposing existing drugs. As for cancer drug development, a critical demand for the agents to target low incidence mutation is inevitable.
Although AI/ML can improve the design of preclinical experiments and clinical trials, help to match correct patients with clinical trials and even optimise clinical trials, current regulatory requirements indicate that having enough patients for low incidence events is a big obstacle. Fortunately, AI also provides a solution, which is in silico patients. With AI patients, we can run in silico clinical trials to identify responders and optimise therapy combinations, the line of therapy and treatment sequences.
Challenges and future directions
One of the key challenges in the process of drug development is ensuring drug safety. Translating knowledge on the known effects of drugs to anticipate their side effects is a difficult process. Scientists and engineers from academic institutions and pharmaceutical industries such as Roche and Pfizer have sought to use AI/ML to extract useful knowledge from data gathered in clinical trials. Currently, an active area of research is the analysis of this data in the context of drug safety17.
Yet, despite all the promises of AI/ML technology in precision oncology, obstacles and pitfalls are formidable in the real world of cancer patient care. A recent example is IBM’s AI algorithm, Watson for Oncology18. This algorithm was based on a small number of synthetic cases with very limited real data. Many of the recommendations were shown to be erroneous, such as suggesting the use of bevacizumab in a patient with severe bleeding – which represents an explicit contraindication for the drug.
However, the efficiency and usefulness of AI/ML algorithms depends heavily on the accuracy and consistency of the data they are trained on. This doesn’t mean the failure of AI in clinical care. In fact, various algorithms essentially share similar principles and no revolutionary algorithm method has emerged so far.
More comprehensive and accurate training data is the key to the success of AI in precision medicine. Therefore, greater effort should be put into generating data and collecting data from diverse populations. Exactly how this revolutionary role of AI will improve the real clinical world remains to be demonstrated and will be dependent on the availability of comprehensive, trustworthy and diverse patient data19.
To this end, crowd-source challenges have been designed to tackle cancer genomics, using experimental data to objectively and transparently evaluate the accuracy. Collaboration between many research entities and pharmaceutical companies has also enhanced access to a range of candidate drugs and a large number of tumour-specific consortia20,21,22. While such collaborations improve the chances of success, they increase the complexity of precision medicine for heterogenous diseases.
Genomic medicine and ML have been successful in identifying rare diseases with a different frequency of occurrence too (eg. rheumatologic versus rare inborn errors of metabolism). Data in the literature of the last ten years alone contains at least 74 different cases of rare diseases23. To this end, a possible route could be establishing a reward for the pharmaceutical industries for investing in medicines that target a small number of patients with rare diseases. Nevertheless, there are encouraging examples of pharmaceutical companies supporting rare diseases. Novartis collaborated with the private company Pharming to work on a molecule for the treatment of Activated PI3K delta syndrome (APDS), an ultra-rare autoimmune disease. Similarly, GlaxoSmithKline expressed interest in the treatment of the same pathology and is testing the use of a drug administered by inhalation.
Rare diseases contribute to a significant proportion of morbidity and mortality in populations with high rates of consanguinity, including in Arab populations. They are estimated to be the second leading cause of infant mortality in Qatar. Many genetic diseases that plague Arab populations are not greatly shared with other populations; therefore, genetic testing developed for other populations are of limited value for Arab communities. Qatar is becoming a precision medicine hub for rare diseases, and we expect to see more biotech and pharmaceuticals companies setting-up joint business ventures here in the coming years.
The ways in which AI/ML can support precision medicine are truly innovative and could certainly lead to major scientific achievements. Precision medicine is already bringing significant improvement in rare diseases, with previously undiagnosed diseases being identified, and patients with potentially lethal cancers are now seeing higher life expectancies and, in some cases, cures.
AI is evolving by itself but with all these trials generating big data, it will also foster the evolution of AI further. Human populations have been combating chronic diseases such as diabetes, obesity, cancer and rare diseases for the last several decades. Will AI prove to be the potent weapon that will turn the tide in this long-ranging battle? Many countries and stakeholders are betting on it.
In high income countries, AI powered healthcare practices have already been put into place. For example, in the UK and Singapore, national AI-based initiatives have been introduced to effectively deal with the burden of many diseases. Qatar is joining in this battle by launching national initiatives to implement precision medicine powered with AI. The country is on its way to becoming a hub of experimentation and hopes to lead neighbouring countries in the Middle East and North Africa in this most important endeavour24.
In addition to Dr. Chouchane and Dr. Sheikh, the following people contributed to this article: Murugan Subramanian, AtilioReyes Romero and Jingxuan Shan from the Genetic Intelligence Laboratory, Weill Cornell Medicine-Qatar.
The authors would like to thank Qatar National Research Fund and the World Innovation Summit for Health (WISH) for their support.
Volume 22, Issue 1 – Winter 2020/21
About the authors
Dr Lotfi Chouchane is a professor in the departments of genetic medicine, and microbiology and immunology at Weill Cornell Medicine-Qatar. He received his Ph.D in immunology from the Pasteur Institute of Paris and the University of Paris VII. He also holds a D.Sc in human genetics and immunology.
Dr Javaid Sheikh is the Dean of Weill Cornell Medicine-Qatar. He joined WCM-Q as Vice Dean for Research and Professor of Psychiatry in 2007 from Stanford University School of Medicine, where he was an Associate Dean and Professor of Psychiatry and Behavioral Sciences
- Melamud, E., et al., The promise and reality of therapeutic discovery from large cohorts.J Clin Invest, 2020. 130(2): p. 575-581.
- Doing well? Fullfilling the promise of precision medicine. The Economist Intelligence Unit 2020. (https://www.eiu.com/n/fulfilling-the-promise-of-precision-medicine/)
- Ho, D., et al., Enabling Technologies for Personalized and Precision Medicine.Trends Biotechnol, 2020. 38(5): p. 497-518.
- Sudlow, C., et al., UK biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age.PLoS Med, 2015. 12(3): p. e1001779.
- All of Us Research Program, I., et al., The “All of Us” Research Program.N Engl J Med, 2019. 381(7): p. 668-676.
- Wang, M., et al., Associations between stressful life events and diabetes: Findings from the China Kadoorie Biobank study of 500,000 adults.J Diabetes Investig, 2019. 10(5): p. 1215-1222.
- Al Thani, A., et al., Qatar Biobank Cohort Study: Study Design and First Results.Am J Epidemiol, 2019. 188(8): p. 1420-1433.
- Fakhro, K.A., et al., The Qatar genome: a population-specific tool for precision medicine in the Middle East.Human Genome Variation, 2016. 3(1): p. 16016.
- DiMasi, J.A., H.G. Grabowski, and R.W. Hansen, Innovation in the pharmaceutical industry: New estimates of R&D costs.J Health Econ, 2016. 47: p. 20-33.
- Vamathevan, J., et al., Applications of machine learning in drug discovery and development.Nat Rev Drug Discov, 2019. 18(6): p. 463-477.
- Ning, W., et al., Open resource of clinical data from patients with pneumonia for the prediction of COVID-19 outcomes via deep learning.Nature Biomedical Engineering, 2020.
- Zeng, X., et al., Target identification among known drugs by deep learning from heterogeneous networks.Chemical Science, 2020. 11(7): p. 1775-1797.
- Tsou, L.K., et al., Comparative study between deep learning and QSAR classifications for TNBC inhibitors and novel GPCR agonist discovery.Sci Rep, 2020. 10(1): p. 16771.
- Koscielny, G., et al., Open Targets: a platform for therapeutic target identification and validation.Nucleic Acids Res, 2017. 45(D1): p. D985-D994.
- Azuaje, F., Artificial intelligence for precision oncology: beyond patient stratification.npj Precision Oncology, 2019. 3(1): p. 6.
- Ho, D., Artificial intelligence in cancer therapy.Science, 2020. 367(6481): p. 982-983.
- Ekins, S., et al., Exploiting machine learning for end-to-end drug discovery and development.Nat Mater, 2019. 18(5): p. 435-441.
- Miller, A., The future of health care could be elementary with Watson.CMAJ, 2013. 185(9): p. E367-8.
- Wilkinson, J., et al., Time to reality check the promises of machine learning-powered precision medicine.The Lancet Digital Health, 2020. 2(12): p. e677-e680.
- Zardavas, D., et al., The AURORA initiative for metastatic breast cancer.Br J Cancer, 2014. 111(10): p. 1881-7.
- Khan, S.S., A.P. Chen, and N. Takebe, Impact of NCI-MATCH: a Nationwide Oncology Precision Medicine Trial.Expert Review of Precision Medicine and Drug Development, 2019. 4(4): p. 251-258.
- Shah, R. and B. Khan, The MATRIX trial.Lancet, 2019. 393(10183): p. 1803.
- Schaefer, J., et al., The use of machine learning in rare diseases: a scoping review.Orphanet J Rare Dis, 2020. 15(1): p. 145.
- Qoronfleh, M.W., et al., THE FUTURE OF MEDICINE, healthcare innovation through precision medicine: policy case study of Qatar.Life Sciences, Society and Policy, 2020. 16(1): p. 12.