Machine learning identifies early predictors of type 1 diabetes

A recent study published in the journal cell report medicine Scientists used plasma protein proteomics to identify proteins associated with the development of type 1 diabetes, according to the Journal.

More than 2,250 samples from 184 participants yielded 376 regulatory proteins identified using machine learning analysis predictive of autoimmunity that precedes type 1 diabetes.

These results provide insight into the pathways altered during the development of type 1 diabetes, allowing disease to be predicted 6 months before onset.

study: Plasma protein biomarkers predict the development of persistent autoantibodies and type 1 diabetes 6 months before the onset of autoimmunity. Image credit: OleksandrNagaiets/Shutterstock.com

What is type 1 diabetes?

Type 1 diabetes (T1D) is an autoimmune disease that affects an estimated 20 million people worldwide, reducing their life expectancy by 11 years. It is characterized by the body’s rejection and destruction of β-cells due to the development of autoantibodies against an individual’s islet proteins, a process called ‘seroconversion’. A cure for this condition does not yet exist.

β-cells are responsible for insulin production, and their disruption leads to many diseases such as blindness, renal failure, and cardiovascular disease. So far, the triggers and mechanisms of T1D are poorly understood.

Recent programs, including the Environmental Determinants of Diabetes of the Young (TEDDY) study, have been launched to elucidate T1D and enable future therapeutic interventions.

These programs have identified plasma proteomics as a viable means to identify biomarkers associated with T1D, thereby gaining insight into the genetic and environmental determinants of disease.

Analysis of these proteins may improve the predictive power of researchers and, in the future, provide healthcare workers with a viable avenue to treat T1D. Unfortunately, many previous studies have failed to systematically validate study participants, leading to confusion in the interpretation of results.

About research

In this study, researchers conducted a nested case-control study of individuals in the TEDDY cohort. The two-phase study was divided into a discovery phase followed by a validation phase.

In the discovery phase, 184 randomly selected donors (92 samples + 92 controls) aged 0–6 years each provided 2,252 plasma samples collected at multiple time points over 18 months. These samples were sequenced using mass spectrometry and the resulting proteome was analyzed to identify the 14 most abundant proteins in each sample.

The validation phase consisted of 990 donors specifically selected based on biomarkers, genetic and demographic characteristics. The researchers developed and implemented a quality control analysis on real-time (QC-ART) system to ensure the quality of data collection and automated data management over the 18-month study.

Thus, 36,252 peptides were identified from 1,720 proteins, of which the 376 most frequently repeated proteins with the highest coefficient of variance were used for statistical analysis.

The researchers ultimately utilized machine learning (ML) models to predict phenotypes based on the 376 proteins identified in phases 1 and 2.

The model specifically tested whether the identified proteins could serve as biomarkers to predict whether a donor would remain in the islet autoimmune (IA) stage or progress to T1D. We ran 100 bootstrap iterations of these models and used logistic regression with a lasso penalty to build and identify the best model.

research result

In this study, we identified 376 proteins associated with the spectrum of IA, from normoglycemia to full T1D development.

These proteins were overexpressed in coagulation and complement cascade-related processes known to co-occur with T1D-related nutrient digestion and absorption, inflammatory signaling, blood clotting, and cellular metabolism.

Proteins identified from donors aged 3-9 months were found to successfully predict the development of T1D by age 6 years. A change in protein composition before seroconversion was observed in the donor’s metabolic profile, and the ML model allowed him to predict T1D 6–12 months before disease onset.

This study identified and validated 83 biomarkers that could be used in future clinical studies to identify T1D in patients with a genetic predisposition to T1D.

We believe that evaluation of these promising predictive protein panels in other ongoing prospective studies of autoimmunity and T1D development in human cohorts may aid in prognostic and therapeutic development. . ”

A major limitation of this study was that all donors were from the TEDDY study cohort, ie, individuals with a genetic predisposition to T1D and of American and European descent. Further studies involving people from more diverse geographic areas and those without a family history of T1D would help improve the robustness of these results.

Conclusion

Using thousands of TEDDY donor samples, researchers identified 376 proteins associated with future development of type 1 diabetes.

Machine learning models use these proteins to determine whether individuals with different permutations of these proteins remain carriers of T1D or develop autoimmune disease up to 6 months before disease onset. seroconversion may be accurately predicted.

Of the identified proteins, 83 have been termed ‘biomarkers’ and could be used in future clinical and scientific trials. This study is a robustly validated first step in understanding the underlying genetic mechanisms and environmental triggers of T1D.

This lays the groundwork for future studies that build on more geographically diverse samples. Ultimately, this study may pave the way for previously unavailable therapeutic interventions for this widespread condition.