The facts and fictions of precision medicine in autoimmune diseases: is the machine learning approach the response?

In the last decade, the growth of big data and the enormous advance in the discovery of genomic, transcriptomic and proteomic patient profiles have opened new avenues in the field of precision medicine. Application of complex statistical models and availability of engineering technologies have improved our understanding of disease mechanisms and allowed earlier diagnosis and the identification of more effective and tailored treatment according to the unique individual profile, ultimately promising disease prevention. In this setting, systemic autoimmune diseases provide an excellent field for precision medicine model application due to their high clinical heterogeneity and great variability in disease phenotype, progression and treatment response [1].

In their recent paper, Toro-Domínguez and Alarcón-Riquelme [2] deeply reviewed current advances in the research of molecular biomarkers potentially promoting the transition to precision medicine and of tissue molecular and cellular patterns that may be employed to identify the best tailored therapy for each patient. In particular, type I IFN signature, synovial tissue transcriptomic and single cell analyses in rheumatoid arthritis (RA) patients have been highlighted as novel promising tools for the identification of specific pathways that may be relevant to drive therapy for a specific patient or patient subset [2]. In recent years, the type I IFN pathway has emerged as a relevant marker in the immunopathology of systemic autoimmune diseases, in particular systemic lupus erythematosus (SLE), and a potential target for the development of novel tailored therapies. Indeed, it is increasingly recognized as a shared signature pathway that allows the identification of a common serologic and haematologic profile in patients with different systemic autoimmune diseases as well as a specific disease cluster in Sjogren's syndrome (SS) patients characterized by heightened expression of the IFN gene signature [3, 4].

An additional relevant issue relies on the intriguing application of cellular, molecular and transcriptional profiling and single-cell RNA sequencing of synovial tissue in the identification of RA patient clusters with the higher probability of benefitting from a specific treatment. In this setting, the investigation of gene expression profiling and cellular and molecular immunopathology of synovial membrane in RA, salivary gland in SS, kidney in SLE and skin in systemic sclerosis may offer unique information to improve early diagnosis of undifferentiated forms, to stratify patient phenotype and to guide the best targeted therapy by prediction of response to treatment [5]. Moreover, genetic background and molecular disease markers may be strongly influenced by a complex interaction of epigenetic mechanisms and nutritional elements. Lifestyle habits, microbiota composition and consumption of specific foods may exert different effects on disease outcome and treatment response, suggesting that individualized diet composition and specific nutritional elements should be considered as components of personalized medicine [6]. In this setting, multidisciplinary studies aiming to investigate the effects of specific and personalized diet habits on treatment response and disease outcome in patients with autoimmune diseases should be strongly encouraged [7].

In this scenario, big data, with its large number of epidemiologic and clinical data based on international registries, probably represents the best tool to collect a huge amount of genetic, epigenetic, proteomic and metabolomic information as well as of potential biomarkers. Moreover, the availability of serum/plasma and tissue biobanks of cohorts of established as well as preclinical systemic autoimmune diseases will provide a unique opportunity to chart the course of a disease and identify factors associated with increased risk of disease development [8]. Notably, an important issue raised by the Authors is the actual possibility of reliable integration and interpretation of such data in the perspective of early disease diagnosis, tailored prognosis or, hopefully, prevention. Indeed, the combination of ‘omic’ data, laboratory biomarkers and clinical features has very limited value without proper methods of interpretation. In this setting, digital health technologies, including artificial intelligence, machine learning techniques and deep neural networks, represent a valuable tool to integrate such large multidimensional data in a wide spectrum of application fields, including identification of patients at risk for autoimmune disease, classification of disease subsets sharing similar molecular pathways, prediction of disease progression and outcome and estimation of treatment response [9]. The last few years have witnessed the enormous expansion of studies that employed artificial intelligence and machine learning models based on genome-wide expression data, DNA methylation signatures or longitudinal clinical data to predict response to a specific therapy or risk of disease flare in tapering treatments in RA patients [10, 11]. In contrast, integration of serologic, epidemiologic and clinical features of different cohorts of SS patients by artificial neural networks have allowed the identification of specific disease clusters characterized by higher risk of cardiovascular comorbidity [12].

Collectively, these data and recent evidence support that integration of large molecular and clinical data from patients with systemic autoimmune diseases by computational modelling techniques represents a valuable tool to be employed in the future for the implementation of early diagnosis, clinical decision making, prognosis estimation and tailored treatment, despite disease heterogeneity in presentation, diagnosis, course and outcome. Hopefully, in the near future, application of these complex models with high prediction accuracy and reproducibility will lead to disease prevention as the ultimate goal of precision medicine [1]. However, it is important to acknowledge intrinsic limitations to this impersonal technological and machine-based approach to the single individual suffering from a chronic disease. Application of machine learning may be hindered by the complexity of each disease, the lack of validation of several biomarkers and, more importantly, by the impossibility of universal access to powerful diagnostic and laboratory technologies. Indeed, each patient has a unique and distinctive phenotype and identified biomarkers are not all generalizable to each patient sample. As for all data analyses, data quality, availability, completeness and accuracy should be widely guaranteed to ensure correct applicability of the results. Finally, and more importantly, data reproducibility in each clinical setting should be a common target to increase patient quality of life, to reach the optimal treatment for each patient for the proper duration and to predict disease outcome. Thus, the ultimate goal of personalized medicine will be the prevention of the disease and consideration of the natural diversity of each patient with a chronic disease.

Funding: No specific funding was received from any bodies in the public, commercial or not-for-profit sectors to carry out the work described in this article.

Disclosure statement: The authors have declared no conflicts of interest.

Data availability statement

Data are available upon reasonable request by any qualified researchers who engage in rigorous, independent scientific research, and will be provided following review and approval of a research proposal and Statistical Analysis Plan (SAP) and execution of a Data Sharing Agreement (DSA). All data relevant to the study are included in the article.

References