AI-based investigation of molecular biomarkers of longevity.

摘要: In this paper, I build deep neural networks of various structures and hyperparameters in order to predict human chronological age based on open-access biochemical indicators their specifications from the NHANES database. total, 1152 are trained tested. The algorithms tested incomplete data: missing values data records extrapolated by mean or median for each parameter. select best terms validation accuracy (coefficient determination absolute error). It turns out that most accurate results delivered multilayer (6 layers) with recurrent layers. Neural network types selected trial error. reached an 78% coefficient 6.5 also list empirically determined features increase task prediction. Obtained can be considered as approximation biological age. Parameters training datasets broadly: all potentially relevant parameters (926) database used. Although data, they demonstrated ability make reasonable predictions (with R2 > 0.7) no more than 100 indicators. Hence, practical reasons full 926 not required, although analysis impact indicator is useful theoretical developments.