Alternative splicing events as peripheral biomarkers for motor learning deficit caused by adverse prenatal...

Area of Science:

• Neuroscience
• Genetics
• Computational Biology

Background:

• Neurobehavioral deficits in children from adverse pregnancies lack accurate predictive biomarkers.
• Maternal alcohol consumption and diabetes during pregnancy can lead to severe neurodevelopmental issues.
• Predicting deficit severity requires reliable biological signatures and identification tools.

Purpose of the Study:

• To identify peripheral biomarkers for predicting motor learning deficits in offspring from adverse pregnancies.
• To evaluate the efficacy of deep learning models in identifying these biomarkers.
• To explore the underlying molecular mechanisms of neurobehavioral deficits.

Main Methods:

• Analysis of alternative splicing (AS) patterns in lymphocyte RNA from mouse models.
• Development and training of a deep-learning model to predict motor learning deficits.
• Application of Shapley-value analysis to interpret the deep-learning model's predictions.
• Gene ontology and structure-function analyses using AlphaFold2.

Main Results:

• Significant changes in lymphocyte RNA AS patterns serve as accurate peripheral biomarkers for motor learning deficits.
• A deep-learning model identified 29 common AS events across prenatal alcohol exposure (PAE) and offspring of mothers with diabetes (OMD) as superior predictors.
• Shapley-value analysis elucidated the contribution of specific AS events to motor learning deficits.
• AS patterns showed opposite directions in PAE and OMD, suggesting differential RNA-binding protein expression.

Conclusions:

• Alternative splicing of lymphocyte RNA is a valuable resource for discovering peripheral biomarkers of neurobehavioral deficits.
• Deep learning provides an effective computational tool for identifying these biomarkers.
• This approach holds potential for diagnosing and managing neurodevelopmental issues in children exposed to diverse adverse pregnancy conditions.