Machine learning for detection of cardiorespiratory alterations as signatures of autoresuscitation failure in neonate mice

Sudden Unexpected Infant Death (SUID), including Sudden Infant Death Syndrome (SIDS), is the most common cause of infant death in the United States, accounting for approximately 3,700 annual deaths of children under 1 year of age. Although the etiology of SIDS is unknown, previous works support the hypothesis that impairment of the neonate autoresuscitation reflex is a common endpoint in SIDS. The autoresuscitation reflex is a protective response to hypercapnia and hypoxia in which an infant undergoes deep gasping to restart the cardio-respiratory system. However, measurable in vivo physiological biomarkers that can be used as a diagnostic tool to identify SIDS-vulnerable infants currently do not exist. Thus, the goal of our project was to combine modern machine learning with novel mouse modeling and assaying capabilities as groundwork for identifying biomarkers for SIDS. We leveraged a new automated phenotyping platform, Looper, to expose over 800 C57Bl6/J mouse pups to repeated hypercapnic anoxia challenges and recovery periods that continue in a loop until death. These challenges are designed to induce the autoresuscitation reflex, which allows us to study cardiorespiratory behavior before, during, and after the reflex is triggered. We then applied modern machine learning techniques to interrogate the large-scale cardiorespiratory data. We applied several supervised learning methods to model nonlinear relationships between variables derived from breath-wise and challenge-wise statistics of respiration and cardiovascular activity, including ventilatory frequencies, tidal volumes, heart rates, and recovery times. The best-performing predictive model was selected using nested cross-validation. A post-hoc leave-one-covariate-out approach was then applied to the chosen model to assess the statistical significance of each cardiorespiratory variable for predicting the number of challenges survived. The findings from our analysis showed several cardiorespiratory features related to the recovery phase of the first anoxic challenge, baseline acclimation period, and temporal latency to the recovery of breathing and heart rate after a challenge that are significantly correlated with the expected number of challenges survived. Using the novel data-driven insights gleaned from our machine learning analysis, we aim to derive mechanistic insights on the underlying neuropathological and physiological systems that underlie the control of breathing in neonates during stress, to test the effect of neuronal manipulations on the relationships between survivability and cardiorespiratory outcomes, and to hypothesize potential diagnostics and therapeutic interventions for SIDS. FUNDING: NIH/NHLBI R01HLN161142-01 and R01HL130249-07 This abstract was presented at the American Physiology Summit 2025 and is only available in HTML format. There is no downloadable file or PDF version. The Physiology editorial board was not involved in the peer review process.