Extended Data Fig. 2: Aging model training and testing.
From: Organ aging signatures in the plasma proteome track health and disease
a, A bagged ensemble of least absolute shrinkage and selection operator (LASSO) aging models was trained for each of 11 major organs using the mutually exclusive organ-enriched proteins identified as inputs. An “organismal” aging model using the 3907 organ-nonspecific proteins and a “conventional” aging model using all 4778 QC’ed proteins on the SomaScan assay were also trained. Models were trained from the 1,398 healthy individuals in the Knight-ADRC cohort. To reduce overfitting, the LASSO regularization parameter α was determined with bootstrap resampling by selecting sparser model α that provided 95% of maximum training set performance. An individual’s predicted age was defined as the average predicted age across all bootstrapped models. The entire model training scheme for a single example aging model is shown. b, Models were tested in four independent cohorts (Covance, LonGenity, Stanford-ADRC, SAMS). Age predictions from a single example aging model across test cohorts is shown.
