Do you or does anyone you know suffer from a rare disease? Then you will surely be interested in finding out more about how the implementation of AI can become a gamechanger in the field of so-called orphan disease research, from diagnostics and monitoring of disease progression to therapeutic strategies and personalised drug development.
I was reminded about an article by Italian researchers titled
The Impact of Artificial Intelligence in the Odyssey of Rare Diseases that I had read a while ago, when I noticed Anil Mankar liking this post:
I am not sure whether there is any connection between this company and Brainchip or whether our co-founder simply wanted to show his appreciation for a bunch of people wanting to “make a meaningful impact in changing the lives of patients living with severe, complex diseases” - for all we know, his life or that of his loved ones or friends could be affected.
Companies such as Probably Genetic may not necessarily need neuromorphic technology when implementing AI in order to help patients suffering from rare diseases, but it would be wonderful if Brainchip was somehow involved in contributing to such a worthy cause.
Emerging machine learning (ML) technologies have the potential to significantly improve the research and treatment of rare diseases, which constitute a vast set of diseases that affect a small proportion of the total population. Artificial Intelligence (AI) algorithms can help to quickly...
www.mdpi.com
The Impact of Artificial Intelligence in the Odyssey of Rare Diseases
by
Anna Visibelli
1,*,†
,
Bianca Roncaglia
1,†,
Ottavia Spiga
1,2,3,*
and
Annalisa Santucci
1,2,3
1
Department of Biotechnology, Chemistry and Pharmacy, University of Siena, 53100 Siena, Italy
2
Competence Center ARTES 4.0, 53100 Siena, Italy
3
SienabioACTIVE—SbA, 53100 Siena, Italy
*
Authors to whom correspondence should be addressed.
†
These authors have contributed equally to this work.
Biomedicines
2023, 11(3), 887;
https://doi.org/10.3390/biomedicines11030887
Received: 7 February 2023 / Revised: 28 February 2023 / Accepted: 8 March 2023 / Published: 13 March 2023
(This article belongs to the Special Issue
Artificial Intelligence in the Detection of Diseases)
Download keyboard_arrow_down
Browse Figure
Versions Notes
Abstract
Emerging machine learning (ML) technologies have the potential to significantly improve the research and treatment of rare diseases, which constitute a vast set of diseases that affect a small proportion of the total population. Artificial Intelligence (AI) algorithms can help to quickly identify patterns and associations that would be difficult or impossible for human analysts to detect. Predictive modeling techniques, such as deep learning, have been used to forecast the progression of rare diseases, enabling the development of more targeted treatments. Moreover, AI has also shown promise in the field of drug development for rare diseases with the identification of subpopulations of patients who may be most likely to respond to a particular drug. This review aims to highlight the achievements of AI algorithms in the study of rare diseases in the past decade and advise researchers on which methods have proven to be most effective. The review will focus on specific rare diseases, as defined by a prevalence rate that does not exceed 1–9/100,000 on Orphanet, and will examine which AI methods have been most successful in their study. We believe this review can guide clinicians and researchers in the successful application of ML in rare diseases.
Keywords:
rare disease;
machine learning;
artificial intelligence;
precision medicine;
data analysis
1. Introduction
The term rare diseases refers to a vast set of diseases that affect a small proportion of the total population; there are more than 7000 known disorders, and an estimated 250 rare new diseases are discovered annually [1]. In addition, diseases with a prevalence of <1 case per 50,000 population are defined as ultra-rare [2]. This definition is related to a prevalence threshold [
3], and it differs depending on the jurisdiction. In Europe, the European Medicines Agency considers a prevalence of less than 5 in 10,000 people (less than 1 in 2000) [
4], while in the United States, diseases affecting less than 200,000 people in the country were defined as rare by the Orphan Drug Act in 1983 [
5]. In Japan, the
Ministry of Health, Labor, and Welfare defines a threshold of fewer than 50,000 individuals in the country (equivalent to less than 1 in 2500 people) [6]. Therefore, an international definition of rare disease is lacking. Although individually they can be considered rare, they collectively afflict more than 500 million people worldwide [7]. Most of these disorders have characteristics that pose serious challenges for both researchers and public health professionals, especially for patients who face not only a loss in terms of health and quality of psychological and social well-being, but also financial burdens [8].
First, the process of diagnosing a rare disease is often long and exhausting. In 25% of patients, it takes between 5 and 30 years after disease onset to receive a correct diagnosis, which requires the participation of a competent and comprehensive clinical team [9]. A survey conducted from October 2019 through March 2020 by the National Organization for Rare Disorders [10] on 1108 individuals found that 50% of patients and caregivers attribute diagnostic delays to a lack of knowledge about the disease, while 42% believe that delays are caused by limited medical specialization. Many patients identified the problem of doctors not being able to link symptoms, particularly between different organ systems, in addition to the fact that waiting times to consult specialists are long and there would be a need for more tests. A diagnostic delay can have tremendous effects on the patient’s clinical picture, so prompt and accurate diagnosis are the starting point for being able to find therapeutic interventions and resources that can ensure a good clinical outcome [11].
What complicates the rare disease odyssey is that the diagnosis is never the end of the journey, since even from a prognostic and therapeutic point of view, there are huge gaps to be filled [12]. The difficulties encountered at the prognostic level are related to the lack of valid parameters and/or biomarkers, since the molecular pathophysiological mechanisms are still largely unknown. Moreover, the small number of patients does not allow statistically significant parameters to be derived [13]. Thus, the prognosis of patients may change depending on various genetic and environmental factors, but it is complicated to arrive at standards of care for treatment and rehabilitation because health research is necessarily conducted on a small scale and cannot be based on evidence or experience [14]. Conventionally, it takes 10 to 15 years to bring a drug to market, with an average R&D cost of $2.6 billion [15]. These two factors represent a bottleneck in the drug discovery pipeline for rare disorders, as research costs are high while revenues are low due to the small number of patients. This implies that the development of new drugs and treatments can be time-consuming and hindered by the lack of data and funding [16]. The heterogeneous patient populations, often unknown etiology and pathogenesis, the timing of disease progression, and the lack of exhaustive clinical studies make the search for specific drugs very difficult [17]. The key problems related to the development of drugs and therapies for rare diseases are [
13]:
- only small cohorts of patients are interested in purchasing these drugs, making them so-called orphan drugs because they are not competitive for pharmaceutical companies.
- difficulties in treatment because most rare diseases are caused by genetic errors and/or have a degenerative nature.
- significant percentages of patients do not respond to available therapies due to partial or complete loss of response.
Therefore, rare diseases are often referred to as orphans as they fail to attract political, financial, and research interests, even though laws have been passed over the years to address this problem; the US Orphan Drug Act in 1983 and the European Union Regulation on Orphan Medicines in 2000 have rewarded innovation and focused on the value of healthcare for rare disease patients.
Nevertheless, for most of them, there are no adequate therapeutic options. Over the years, specialized interdisciplinary centers for rare disorders have been established, where doctors and researchers can exchange opinions and ideas, creating networks of knowledge and experience that can help patients [
9].
Possible innovative answers to biomedical and clinical challenges come from the world of information technology, and a striking example has been the fight against COVID-19. Since the beginning of the pandemic, artificial intelligence (AI) has played a crucial role in the battle against the virus, and several methods have been applied for various purposes [
18]. Machine learning (ML) and deep learning (DL) models have been used for the early detection and diagnosis of COVID-19 by monitoring the demographic, clinical, and epidemiological characteristics of patients, and for developing diagnostic tools that can quickly analyze CT scans and X-rays to identify patterns indicative of the disease [
19]. AI has also been used to predict patient vulnerability, in order to administer appropriate drugs and treatments [
20], as well as being decisive in accelerating the discovery of potential vaccines. Similarly, AI has been an essential ally for public health policies in contact tracing, monitoring the spread of the virus, and creating predictive models that have helped to identify potential outbreaks. Thus, it is clear how AI is increasingly coming to the aid of physicians at every stage of disease management, to evaluate the efficacy of medical treatments or deeply investigate the correlation between patients and treatments according to their own molecular characteristics.
The precision medicine approach is widely applied to the healthcare area, in particular to rare diseases with the creation of patient registries leveraging large amounts of data to discover potential links. It is a comprehensive and prospective approach to prevention, diagnosis, treatment, and monitoring, built on the genetic characteristics of the individual. Harmonizing databases and including registries are the major facilitators to understand the complexity of diseases, to conduct clinical trials, to improve the drug development process, and to assign the right treatment to the right individual after reliable patient stratification. AI is an ally that can integrate and analyze heterogeneous data (e.g., multi-omics data as well as images). However, first-generation AI systems, which rely on the development of algorithms for diagnosis and treatment that are trained on big data, are not always adequate to meet the needs of rare diseases. Data scarcity and sparsity characterize these disorders, due to fragmented knowledge and the limited number of data and specimens available [
21]. Phenotype and disease severity as well as pharmacogenomic and pharmacokinetic factors are the elements on which successful diagnosis and treatment depend [
13]. Diagnostic decision support systems (DDSS) already exist, i.e., expert systems that support doctors in facilitating the diagnostic process by incorporating medical knowledge. These systems have been proven effective and have improved clinical diagnosis by compiling lists of appropriate differential diagnoses for a given sample of tests [
22,
23]. For rare diseases, these systems need to be implemented.
Networks and registries have been built to bring together data and expertise on rare diseases, making them free, accessible, and shareable worldwide. One example is Orphanet, which over the past 20 years has become the go-to source for information on rare diseases, facilitating access to information and means to identify potential patients, and contributing to the development and sharing of knowledge. Other available datasets are the Online Mendelian Inheritance in Man and Human Phenotype Ontology. The knowledge deposited in these databases is used by DDSSs built for rare disorders, some examples of which are FindZebra [
24], PhenoTips [
25], Rare Disease Discovery [
26], and Ada DX [
27].
Second-generation AI systems were designed to fill the diagnostic, prognostic, and therapeutic gaps that must be overcome to achieve patient-centricity for patients with rare diseases [28] (Figure 1). These systems use a personalized closed-loop system designed to enhance end-organ function, overcome problems of tolerance or loss of efficacy, and improve patients’ responses to chronic drugs [13] in a precision medicine perspective.
Figure 1. Examples of ML applications within diagnosis, prognosis, and treatment.
(…)
5. Conclusions
ML methods have shown promise in the identification and diagnosis of rare diseases. With the vast amounts of data now available through electronic health records and heterogeneous databases, AI algorithms can help quickly identify patterns and associations that would be difficult or impossible for human analysts to detect. For example, researchers use ML algorithms to analyze patient data and identify characteristic patterns associated with certain rare diseases. By doing so, these tools can help to narrow down the list of possible diagnoses, making it more likely that patients will receive a correct diagnosis in a timely manner.
Predictive modeling techniques, such as DL, have been used to forecast the progression of rare diseases, allowing for earlier interventions and better treatment planning. This could potentially lead to a more accurate classification of rare diseases and enable the development of more targeted treatments. From a precision medicine perspective, by identifying biomarkers associated with a particular rare disease, AI algorithms can help to develop personalized treatment plans, helping to improve patient outcomes and reduce the risk of side effects.
AI has also shown promise in the field of drug development for rare diseases. AI algorithms can be used to analyze patient data and identify subpopulations of patients who may be most likely to respond to a particular drug. This can help to make clinical trials more efficient and increase the chances of a drug being approved for use in patients with rare disorders. Despite the potential benefits of ML, there are still challenges that must be overcome to fully realize its potential in the field of rare diseases. These include a lack of large, well-annotated datasets, and the need for interpretable models that can be easily understood and trusted by clinicians. Rare diseases usually present an unusually big data regime, which is characterized by huge omics data but a limited number of patients. The rarity of orphan patients, despite the presence of registries, still has a large impact on ML analyses, and thus open data can contribute significantly to support the modeling attempt.
In the future, patient registries and open data need to be integrated, translating the largest amounts of data available into potential connections. Finally, a current limitation is the lack of interpretability, which makes it difficult for clinicians and researchers to understand algorithm outputs. In fact, explainability is one of the most debated topics for the application of AI in healthcare. While AI-based systems have been shown to outperform humans in certain analytical tasks, the lack of explainability continues to attract criticism.
In this context, explainable AI approaches are the new frontier of ML applications in healthcare, in order to ensure the understanding, by both clinicians and patients, of the “mental process” followed by the artificial brain to reach a certain decision.
In conclusion, the application of ML techniques can greatly assist rare disease research and treatment, but to use it effectively, it needs to be implemented under the right ethical principles, avoid biases, and also be transparent for the patient. To fully tap into its potential, AI needs to be validated through clinical trials and real-world evidence. Furthermore, it needs to be accompanied by regulatory frameworks that ensure the safety and reliability of AI-based medical devices and diagnostic tools. More work is needed to overcome data-related challenges, ensure fair and trustworthy models, and help translate the research findings into practical applications that can benefit patients and their families.