Censoring-aware Gaussian process modeling identifies time informative functional dynamics associated with spoilage status of modified atmosphere packaged broiler meat

Censoring-aware Gaussian process modeling of broiler meat metatranscriptomes at 4°C predicts sampling time within 0.43 days on average and matches sensory spoilage phases.

• Why it matters: Routine shelf-life assessment is retrospective and based on culturing or sensory readouts, leaving functional metabolic dynamics of the spoilage microbiome only indirectly connected to deterioration rate.
• What they did: The CAGP framework models longitudinal pathway expression while treating low/zero pathway values as left-censored below a detection threshold, using odor-based fresh/early/late spoilage phase labels from storage-day metatranscriptomes at 4°C and 6°C.
• The result: Leave-one-out prediction within the 4°C series yields mean absolute errors of 0.43 days (leave-one-replicate-out) and 0.78 days (leave-one-day-out), and trajectories learned at 4°C transfer to 6°C at the spoilage-phase level.

AI summary of the abstract · may contain errors · not medical advice · read the paper.