The invention discloses a TabNet-based interpretable multi-tissue
DNA methylation age prediction method and
system, and relates to the technical field of
bioinformatics and computer crossing, and the method comprises the following steps: obtaining
DNA methylation data of a to-be-detected sample, and preprocessing the data; a
methylation beta value of a
feature set composed of 2035 preset CpG sites is extracted from the preprocessed data to serve as an input feature; the input features are loaded to an
age prediction model based on a TabNet framework obtained through a specific training process for calculation, the predicted
DNA methylation age of the to-be-detected sample is output, and through screening of a simple preset
CpG site with multi-tissue stability,
tissue specific signals are accurately captured in cooperation with a TabNet dynamic attention mechanism, so that the
DNA methylation age of the to-be-detected sample is predicted. The
black box dilemma of other
deep learning models is effectively solved, and meanwhile, the technical problems of insufficient prediction precision, weak generalization ability in extreme
age groups, poor robustness in practical application and the like of a traditional
machine learning model are solved.