This research summary article is based on the research paper 'Progression of type 1 diabetes from latency to symptomatic disease is predicted by distinct autoimmune trajectories' and IBM article 'IBM continues advancing disease progression modeling and biomarkers research using the latest in AI'
Type 1 Diabetes is a kind of autoimmune diabetes that can affect both children and adults. It can cause long-term consequences such as renal failure, heart attack, stroke, blindness, and amputation for people who are affected.
This disorder has neither a therapy nor a preventative, and the incidence of Type 1 Diabetes (T1D) has been rapidly rising in recent decades, necessitating further study into prevention and early detection.
Last year, IBM Research highlighted prior relevant research1 undertaken in conjunction with JDRF and five academic research sites that make up the T1DI Study Group: DAISY, DEW-IT in the United States, DiPiS in Sweden, DIPP in Finland, and BABYDIAB/BABYDIET in Germany. That research advanced our understanding of the development of biomarkers linked to the risk of T1D onset in children, and we discovered that the number of islet autoantibodies present at seroconversion, the earliest stage of autoimmunity development, can accurately predict the risk of T1D onset in young children for up to 10 to 15 years.
The beginning of type 1 diabetes in children is preceded by the development of islet autoimmunity; however, autoantibodies do not always imply visible illness, and the start of clinical symptoms is difficult to predict. They show that illness development follows unique trajectories using longitudinal sampling of islet autoantibodies (IAb) against insulin, glutamic acid decarboxylase, and islet antigen-2. Of the 24662 patients in the combined Type 1 Data Intelligence cohort, 2172 meet the requirements of two or more follow-up visits and IAb positive at least once, with 652 developing type 1 diabetes during the 15-year research.
IBM Research has achieved yet another significant milestone in this field of study. The T1DI Study Group presented new data this week in Nature Communications that reveals that unique autoimmune trajectories can predict the course of Type 1 diabetes from the formation of islet autoantibodies to symptomatic illness. Researchers introduced a novel data visualization tool – DPVis – to IBM Research’s AI and machine-learning technologies for illness progression modeling in this study (DPM Tools). This has allowed us to get wholly new insights from the trial data, which may eventually help us understand the role of islet autoantibodies in the development of T1D, enhancing our capacity to anticipate disease start.
Multiple islet autoantibodies during seroconversion raise the risk of T1D, as IBM previously demonstrated. Still, they may not occur consistently throughout time — and a patient may have various combinations of antibodies at different times. Their earlier study revealed that these changes’ ramifications were unknown, so they decided to investigate this by looking at the complex patterns of antibodies that develop throughout time rather than at a single point in time. They discovered three separate trajectories or “pathways,” each of numerous discrete states linked with various degrees of risk.
By providing a shared framework for their increasing understanding of how biomarkers impact a patient’s journey toward illness onset, AI and data visualizations make it feasible to keep researchers in the loop for this large-scale, long-term cooperation.
Unfortunately, many people nowadays aren’t identified with diabetes until they’ve developed diabetic ketoacidosis, a life-threatening illness with long-term consequences. This discovery may one day make it simpler to identify at-risk youngsters whose families can learn about T1D symptoms, allowing for early diagnosis. At-risk children can also participate in clinical trials to delay or even prevent the onset of T1D.
Efforts to simulate disease development and find biomarkers should be expanded.
The T1D research is part of IBM Research’s larger objective to build AI tools like the DPM Tools to accelerate scientific findings in healthcare and life sciences. They worked with other charities, like the CHDI Foundation for Huntington’s disease, in addition to JDRF, to bring in intense devotion and scientific competence.
Huntington’s disease symptoms usually appear between 30 and 50 and increase over time, eventually leading to severe impairment. While no treatment will stop the disease from progressing, some drugs can help with specific symptoms. Unfortunately, the majority of them include side effects that might have a detrimental influence on individuals with Huntington’s disease’s quality of life.
For numerous years, CHDI and IBM have collaborated on research with academic institutions, addressing various research problems in disease progression modeling, brain imaging, and molecular modeling.
Looking forward to the future
Modern science is a collaborative endeavor, and nowhere is this more evident than in healthcare, where discoveries in illness knowledge and therapy development necessitate collaboration across research teams from all disciplines.
IBM has been actively convening, coordinating, or participating in such multidisciplinary teams, which has resulted in clinically significant results and high-impact journal articles. The three initiatives mentioned here are excellent examples of our approach to scientific discovery through a partnership with discovery communities. We’ve created an ecosystem of reusable methods, models, and datasets due to our work, which is currently being used to investigate other illnesses and will be utilized to widen the scope of our work in the future.
Continuous-Time Hidden Markov Models (CT-HMMs) were learned as disease progression models (DPMs) from the longitudinal T1DI study cohort depending on the presence or absence of IAb. A model incorporating 11 latent states was created using machine learning approaches that best suited the observed data and were then applied to all IAb positive individuals to obtain the conclusions. The revealed states generated three trajectories for the 643 diagnosed (D) individuals, TR1, TR2, and TR3, each defined by a unique sequence of latent states (Fig. a). These were further investigated via interactive data visualization statistical analysis.
Each latent state is represented in this diagram by a set of probabilities for the presence of each IAb (Fig. a). The model was used to identify longitudinal observation sequences of people with IAb positive. These participants were then classified into those who acquired type 1 diabetes during the research period (Diagnosed/D) and those who did not or were lost to follow-up (Undiagnosed/UD). Statistical analysis connected these latent states and their trajectories with other research characteristics to reach conclusions about these two groups from all subjects with IAb positive.
Prathamesh Ingle is a Mechanical Engineer and works as a Data Analyst. He is also an AI practitioner and certified Data Scientist with an interest in applications of AI. He is enthusiastic about exploring new technologies and advancements with their real-life applications