(A) Diagram of data processing pipeline. Three-min raw data from each run are fed through a quality control pipeline to remove data from broken and/or miscalibrated sensors. Data is then analyzed by a robust hidden Markov model (HMM) to assign each 3 min window to one of six states. Gas measurements are then adjusted for body mass. Data are then aggregated in a variety of ways for feature extraction: by time window (1 or 4 hr windows), by periods of behavior (e.g., number of sleeping periods), as a ratio of values from before and after a light transition (e.g., RQ pre-lights on/RQ post-lights on), and by state (e.g., VCO2 while running). Finally, data are used for modeling and analysis. (B) Effect of age on six of the base features of the phenotyping cages. Data were averaged for each run and then averaged across runs within each age bin. (C) Average percent time spent in each state for young (3–6 months) and old (30–33 months) animals, split by dark/light phase. (D) Average occupancy of each HMM state at each hour of the day. Timepoints represent the preceding hour, for example, the 7 pm timepoint includes data from 6 to 7 pm. (E) Effect of age on average duration and daily number of sleeping periods. (F) Effect of age on average duration and daily number of feeding periods. (G) Heatmap of aging trajectories. Rows are features (n = 309), columns are monthly age bins. Each feature was normalized and scaled across age bins from lowest value (0) to highest value (1). (H) Examples of four features that decrease with age, with different ages of onset. All error bars are SEM.