Unleashing the Potential of AI: Revolutionizing Rare Disease Research and Drug Development Through Diversity and Innovation

Deepti Dubey, PhD and Harsha Rajasimha, PhD -IndoUSrare.

Rare diseases present substantial challenges in terms of diagnosis and care navigation. For 25% of patients, obtaining an accurate diagnosis can take an average of five to seven years from the onset of the disease, necessitating the involvement of a proficient and comprehensive clinical team. What further complicates the journey with rare diseases is that the diagnosis marks not the culmination but rather the commencement of the odyssey. From a prognostic and therapeutic standpoint, there exist considerable gaps yet to be bridged. The challenges at the prognostic level stem from the dearth of reliable parameters and/or biomarkers, as the molecular pathophysiological mechanisms remain largely elusive. Additionally, the limited number of patients data points with a given rare disease hinders the derivation of statistically significant parameters. Conventionally, bringing a drug to market takes 10 to 15 years, with an average research and development cost of $2.6B.¹ These factors pose a bottleneck in the drug discovery process for rare disorders; research expenses are high while revenues remain low due to the small patient population. This consequently leads to the protraction of new drug and treatment development, compounded by insufficient data and funding.

Artificial Intelligence (AI)-driven tools have the potential to streamline the research and development process for orphan therapies for rare diseases, reducing both the time and costs by accelerating the timelines, minimizing manual repetitive tasks, and make sense of complex variety of data. In recent years, numerous AI systems utilizing Machine Learning (ML) and Deep Learning (DL) algorithms have been developed to address the diagnostic, prognostic, and therapeutic gaps essential for achieving patient-centric care for individuals with rare diseases.²

AI-Driven Diagnostic Tools

The accurate diagnosis of rare diseases holds crucial importance in patient triage, risk stratification, and targeted therapies. Due to their infrequency, symptoms of rare diseases often present as unfamiliar and atypical to clinicians. This increases the likelihood of patients not receiving an appropriate diagnosis and subsequently missing out on successful therapy. The variability inherent in rare diseases further compounds the challenge of timely identification, owing to the limited accessibility of clinical diagnostic procedures.

A standard diagnostic approach for rare diseases entails a comprehensive assessment of medical history, physical examination, and genetic testing, which may unveil specific mutations associated with the condition. Additionally, imaging studies like X-rays, MRI, or CT scans may be employed. In this context, AI emerges as a potential game-changer, albeit a complex one. Through the development of ML algorithms capable of scrutinizing vast datasets, AI can discern patterns and markers characteristic of specific rare diseases. ML and DL models have demonstrated efficacy in aiding diagnostic decisions based on phenotypic characterization.² Knowledge graphs, leveraging historical data, medical knowledge, and genetic tests, have been widely utilized for disease classification.³ DL-based approaches have been instrumental in gauging disease severity using pathological features such as gait analysis in conditions like Huntington Disease (HD).⁴ Multiple studies have successfully employed ML and DL techniques to differentiate disorders with overlapping clinical manifestations, like Parkinson’s disease and multiple system atrophy, utilizing MRI, CT scans, or X-rays.² ML algorithms, leveraging specific biomarkers or multiomics data, are now at the forefront of early detection efforts for many rare diseases. In a recent study, researchers applied a trained neural network called ConvNetACh, which analyzed heart rate variation data from Rett syndrome patients, effectively distinguishing them from subjects with typical development. This has potential applications as biomarkers for early detection of neurodevelopmental spectrum disorders.⁵ Additionally, DeepMind’s AlphaMissense has made strides in predicting the molecular effects of genetic variants on protein function, contributing to the identification of pathogenic missense mutations and previously unknown disease-causing genes. This development is poised to increase the diagnostic yield for rare genetic diseases.⁶

AI-Driven Prognosis

Most rare conditions are chronic and lifelong, making predictive prognosis crucial for patients. AI can significantly contribute to the prognosis of rare disorders by bridging gaps in data and experience. Through the analysis of extensive datasets, including electronic health records, genomic data, and imaging studies, ML algorithms can discern patterns and forecast outcomes for individuals with rare diseases, offering valuable insights to shape prognoses and guide decisions.

Numerous studies employing ML and DL techniques have identified genetic and protein biomarkers for adrenocortical carcinoma, enabling the prediction of prognosis for this rare and aggressive cancer.⁷ AI aids in comprehending disease progression and predicting survival times using medical data. For instance, researchers utilized immune cell frequency profiles, along with clinical and serological data from patients with juvenile-onset systemic lupus erythematosus (jSLE), to pinpoint predictive disease outcome signatures through AI tools.⁸

AI-Driven Treatment

Nearly 95% of rare diseases lack FDA-approved drug treatments, and the rising number of rare diagnoses places significant pressure on scientists and clinicians to characterize these conditions and align patients with suitable treatments.⁹ With the continuous influx of biomedical data, AI presents an opportunity to convert this knowledge into a usable format for identifying therapeutic strategies. Utilizing ML-based software like Assay Central, researchers are screening compounds in silico before conducting in vitro testing. This approach has proven successful in identifying novel compounds with potential for disease modulation in the treatment of sialidosis.¹⁰

Recently, artificial intelligence (AI) has revolutionized drug toxicity prediction by offering more precise and efficient methods for identifying potentially harmful effects of new compounds before subjecting them to human clinical trials. This not only saves time but also conserves financial resources.¹¹

Lack of Diversity

It has been 22 years since the landmark completion of the draft human genome sequence, resulting in an unprecedented volume of genomic data. This data is scrutinized through genome-wide association study (GWAS)/phenome-wide association study (PheWAS) methods to unveil connections between genotype and phenotype. These discoveries have significantly contributed to pharmacogenomics and enhanced clinical decision support in numerous healthcare systems. However, managing the influx of genomic data from sequencing and clinical information from electronic health records (EHRs) presents formidable challenges for data scientists.

With the emergence of artificial intelligence (AI) technologies like machine learning and deep learning, an increasing number of GWAS/PheWAS studies have successfully harnessed this technology to surmount the challenges.¹² Yet, it is important to note that most genomics studies, including genome-wide association studies (GWAS), have been conducted in individuals of European descent (86.3%), followed by East Asian (5.9%), African (1.1%), South Asian (0.8%), and Hispanic/Latino (0.08%) populations. Data from the International HundredK+ Cohorts Consortium (IHCC), a recently established consortium of international cohort studies, highlights this ancestral disparity, with approximately 22.5 million participants from North America and Europe compared to a mere 0.3 million from South and Southeast Asia (Indian subcontinent).¹³ Significantly, the Indian subcontinent alone constitutes a quarter of the global population, underscoring the potential impact of increased representation in clinical research and genetic databases.

It is imperative to recognize that if training data lacks representation of population diversity, AI may inadvertently perpetuate bias, potentially leading to misdiagnoses in historically underrepresented patient groups. This is exemplified by a case where Face2Gene, an automatic deep-learning algorithm, predicted facial phenotypes of Noonan Syndrome (NS) and Neurofibromatosis type 1 (NF1) syndromes in a Latino-American population with accuracies of 66.7% and 10%, compared to 100% accuracy in the European population.¹⁴ Studies like these underscore the critical importance of incorporating diverse populations in genetic studies and clinical trials to enhance diagnostic methods and therapeutic interventions for rare disorders.

The potential of AI in accelerating rare disease research and drug development is immense. However, this potential is currently compromised by the lack of diversity in the available patient data, participation, and limited focus on specific populations in clinical genomic studies. Recognizing and addressing this challenge is crucial to unleashing the full power of AI in advancing healthcare. This can be accomplished through cross-border collaboration, active engagement with the rare disease patient community, and sharing of resources and knowledge. Organizations like IndoUSrare are instrumental in initiating these cross-border collaborations between the USA and densely populated countries such as India and providing a platform for discussions crucial for rare diseases and orphan drug development. Key highlights and insights from the inaugural Indo US Bridging RARE Summit 2023 can be found at https://www.prweb.com/releases/indo[1]us-bridging-rare-summit-heralds-a-new-era-of-cooperation-for-rare[1]diseases-orphan-drugs-development-301978670.html

References

1. Knowledge-based approaches to drug discovery for rare diseases. Drug Discov Today. 2022 Feb;27(2):490-502.
2. The Impact of Artificial Intelligence in the Odyssey of Rare Diseases. Biomedicines. 2023 Mar 13;11(3):887.
3. Improving rare disease classification using imperfect knowledge graph. BMC Med Inform Decis Mak. 2019 Dec 5;19(Suppl 5):238.
4. A Deep Learning-Based Approach for Gait Analysis in Huntington Disease. Stud Health Technol Inform. 2019 Aug 21;264:477-481.
5. Deep learning of spontaneous arousal fluctuations detects early cholinergic defects across neurodevelopmental mouse models and patients. Proc Natl Acad Sci U S A. 2020 Sep 22;117(38):23298-23303.
6. Accurate proteome-wide missense variant effect prediction with AlphaMissense. Science. 2023 Sep 22;381(6664):eadg7492.
7. Identifying New Potential Biomarkers in Adrenocortical Tumors Based on mRNA Expression Data Using Machine Learning. Cancers (Basel). 2021 Sep 17;13(18):4671.
8. Disease-associated and patient-specific immune cell signatures in juvenile-onset systemic lupus erythematosus: patient stratification using a machine-learning approach. Lancet Rheumatol. 2020 Jul 29;2(8):e485-e496.
9. The Power of Being Counted. RARE-X report June 2022.
10. A New Approach to Drug Repurposing with Two-Stage Prediction, Machine Learning, and Unsupervised Clustering of Gene Expression. OMICS. 2022 Jun;26(6):339-347.
11. Artificial Intelligence in Drug Toxicity Prediction: Recent Advances, Challenges, and Future Perspectives. J Chem Inf Model. 2023 May 8;63(9):2628-2643.
12. How data science and AI-based technologies impact genomics. Singapore Med J. 2023 Jan;64(1):59-66.
13. A roadmap to increase diversity in genomic studies. Nat Med. 2022 Feb;28(2):243-250.
14. Population-specific facial traits and diagnosis accuracy of genetic and rare diseases in an admixed Colombian population. Sci Rep 13, 6869 (2023)

Author Biographies

Deepti Dubey, PhD, is an accomplished researcher with expertise in Molecular Genetics, Neurological Disorders, and Rare Diseases from esteemed institutions. In collaboration with non-profit patient organizations, she has driven research initiatives for accelerating breakthroughs for rare diseases. With a Clinical Trials Specialization, Deepti is a scientific writer at IndoUSrare.

Harsha K Rajasimha, PhD, is a serial entrepreneur, investor, and philanthropist. He serves as the Founder and Executive Chairman of the Indo US Organization for Rare Diseases (https://IndoUSrare.org) with the mission to make rare disease research and clinical trials universally accessible. Harsha is the Founder and CEO of Jeeva Informatics Solutions Inc. (https://jeevatrials.com) an AI-driven unified platform for efficient and universally accessible clinical trial management. Harsha serves as the chairperson of the annual Indo US bridging RARE Summit (https:// summit.indousrare.org)

Publication Detail

This article appeared in American Pharmaceutical Review:
Vol. 26, No. 8
Nov/Dec 2023
Pages: 23-25

Subscribe to our e-Newsletters
Stay up to date with the latest news, articles, and events. Plus, get special
offers from American Pharmaceutical Review delivered to your inbox!
Sign up now!