Leveraging artificial intelligence and machine learning to accelerate discovery of disease-modifying therapies in type 1 diabetes

Abstract

Progress in developing therapies for the maintenance of endogenous insulin secretion in, or the prevention of, type 1 diabetes has been hindered by limited animal models, the length and cost of clinical trials, difficulties in identifying individuals who will progress faster to a clinical diagnosis of type 1 diabetes, and heterogeneous clinical responses in intervention trials. Classic placebo-controlled intervention trials often include monotherapies, broad participant populations and extended follow-up periods focused on clinical endpoints. While this approach remains the 'gold standard' of clinical research, efforts are underway to implement new approaches harnessing the power of artificial intelligence and machine learning to accelerate drug discovery and efficacy testing. Here, we review emerging approaches for repurposing agents used to treat diseases that share pathogenic pathways with type 1 diabetes and selecting synergistic combinations of drugs to maximise therapeutic efficacy. We discuss how emerging multi-omics technologies, including analysis of antigen processing and presentation to adaptive immune cells, may lead to the discovery of novel biomarkers and subsequent translation into antigen-specific immunotherapies. We also discuss the potential for using artificial intelligence to create 'digital twin' models that enable rapid in silico testing of personalised agents as well as dose determination. To conclude, we discuss some limitations of artificial intelligence and machine learning, including issues pertaining to model interpretability and bias, as well as the continued need for validation studies via confirmatory intervention trials.

Keywords: Artificial intelligence; Digital twin; Drug discovery; Drug repurposing; Drug response; Immunotherapy; Machine learning; Pharmacogenetics; Precision medicine; Review; Type 1 diabetes.

Fig. 1

Leveraging AI/ML for smarter trial design. The increasing availability of diverse, high-dimensional datasets and advanced computing technologies is providing opportunities to use AI/ML to model complex phenotypes and biological phenomena at scale. The AI/ML applications depicted here—when informed by domain expertise in type 1 diabetes—are poised to generate substantial gains in the development and translation of disease-modifying therapies for type 1 diabetes. To this end, AI/ML can be effectively leveraged to optimise cohort, therapeutic agent and clinical endpoint selection. T1D, type 1 diabetes. Created with BioRender.com. This figure is available as part of a downloadable slideset

Fig. 2

Application of AI/ML to drug repurposing, development and combination therapies. (a) Drugs that have been previously repurposed from other diseases to type 1 diabetes are shown in blue, including the antigen-presenting cell (APC) regulators golimumab (anti-TNF), abatacept (CTLA-4-Ig) and ustekinumab (anti-IL-12p40); the T cell deplete, ATG (anti-thymocyte globulin); the B cell deplete, rituximab (anti-CD20); the cytokine signalling inhibitor, baricitinib (JAK1/2 blockade); and the Treg enhancer, low-dose IL-2. Drugs that we propose for repurposing in type 1 diabetes are shown in grey, including the cytokine/chemokine signalling inhibitors anifrolumab (anti-IFNAR1), pazopanib (off-target LNK/SH2B3 blockade), deucravacitinib (TYK2 blockade) and maraviroc (CCR5 blockade). (b) Example of ML-powered kinase profiling prediction showing predicted drug–target binding scores. Kinase tree diagram showing that deucravacitinib inhibits TYK2. The kinome-wide inhibitory predict platform (KIPP) identified an off-target effect of deucravacitinib for a GWAS hit related to type 1 diabetes, ITK (IL-2 inducible T cell kinase). Created with KIPP by idrugLab [33]. (c) Opportunities for use of AI/ML algorithms in the design of antigen-specific immunotherapies for type 1 diabetes include models for antigen-to-peptide processing, MHC:peptide binding, QTL analysis of HLA associations, T cell receptor (TCR):peptide binding, and B cell receptor (BCR):antigen binding. PI, proinsulin. (d) Opportunities for AI/ML use in selecting combination therapies include models of drug synergy and drug–drug toxicity. We propose using synergy models to identify factors that can augment in vivo beta cell proliferation beyond that seen with glucagon-like peptide-1 (GLP-1) analogues, dual-specificity tyrosine phosphorylation-regulated kinase 1A inhibitors (DYRK1Ai) and TGF-β receptor inhibitors (TGFβRi). Drug toxicity can occur when drug efflux transporters are inhibited, thereby increasing intracellular concentrations of a second drug. CCR5, C-C motif chemokine receptor 5; IFNAR1, IFN alpha and beta receptor subunit 1; JAK, janus kinase; LNK (also known as SH2B3), SH2B adaptor protein 3; P-gp inh., P-glycoprotein inhibitor; TYK2, tyrosine kinase 2. Created with BioRender.com. This figure is available as part of a downloadable slideset

Fig. 3

Responder identification and potential for AI-enabled precision medicine. (a) Algorithms for predicting drug metabolism and/or availability incorporate information about genetic variants affecting drug metabolism via cytochrome P450 (CYP) enzymes and efflux via ATP-binding cassette (ABC) transporters. While not yet incorporated into drug response algorithms, we propose that (b) SNPs and copy number variants (CNVs) affecting FcγR may aid in predicting response to therapeutic antibodies. (c) HLA-associated development of anti-drug antibodies (ADA), neonatal Fc receptor for IgG (FcRn)/FCGRT variants, and environmental factors including infection and malnutrition regulate drug catabolism, which may also affect drug availability. (d) Together, these variants may inform drug dosing or selection by influencing target depletion, drug metabolism or drug availability. NK, natural killer cell. Created with BioRender.com. This figure is available as part of a downloadable slideset