Development and validation of machine learning-based diagnostic models using blood transcriptomics for early...

Area of Science:

• Genomics and Bioinformatics
• Machine Learning in Healthcare
• Pediatric Endocrinology

Background:

• Early identification of Type 1 Diabetes Mellitus (T1DM) in children is critical for timely interventions.
• Peripheral blood transcriptomic analysis offers a minimally invasive method for early biomarker discovery.
• Predicting T1DM onset in children up to 46 months before clinical diagnosis is the focus.

Purpose of the Study:

• To develop and validate machine learning algorithms for predicting T1DM onset in children.
• To utilize transcriptomic signatures from peripheral blood for early T1DM prediction.
• To identify predictive biomarkers for T1DM in pediatric populations.

Main Methods:

• Analyzed RNA-sequencing data from 247 pre-diabetic children and healthy controls.
• Employed five feature selection methods and nine machine learning algorithms to create 45 model combinations.
• Validated models using quantitative polymerase chain reaction (qPCR) in an independent cohort.

Main Results:

• Significant differential gene expression patterns were observed between pre-diabetic and control groups.
• Four model combinations showed superior predictive performance, accurately predicting T1DM onset up to 46 months prior.
• Elastic Net-based models achieved perfect classification in the validation cohort, indicating clinical viability.

Conclusions:

• Peripheral blood transcriptomics combined with machine learning enables early pediatric T1DM prediction.
• Identified transcriptomic signatures and validated models form a basis for non-invasive diagnostic tools.
• Findings support precision medicine for childhood diabetes prevention, requiring larger cohort validation.