SPINDLE: End-to-end learning from EEG/EMG to extrapolate animal sleep scoring across experimental settings, labs and species

Understanding sleep and its perturbation by environment, mutation, or medication remains a central problem in biomedical research. Its examination in animal models rests on brain state analysis via classification of electroencephalographic (EEG) signatures. Traditionally, these states are classified by trained human experts by visual inspection of raw EEG recordings, which is a laborious task prone to inter-individual variability. Recently, machine learning approaches have been developed to automate this process, but their generalization capabilities are often insufficient, especially across animals from different experimental studies. To address this challenge, we crafted a convolutional neural network-based architecture to produce domain invariant predictions, and furthermore integrated a hidden Markov model to constrain state dynamics based upon known sleep physiology. Our method, which we named SPINDLE (Sleep Phase Identification with Neural networks for Domain-invariant LEearning) was validated using data of four animal cohorts from three independent sleep labs, and achieved average agreement rates of 99%, 98%, 93%, and 97% with scorings from five human experts from different labs, essentially duplicating human capability. It generalized across different genetic mutants, surgery procedures, recording setups and even different species, far exceeding state-of-the-art solutions that we tested in parallel on this task. Moreover, we show that these scored data can be processed for downstream analyzes identical to those from human-scored data, in particular by demonstrating the ability to detect mutation-induced sleep alteration. We provide to the scientific community free usage of SPINDLE and benchmarking datasets as an online server at https://sleeplearning.ethz.ch. Our aim is to catalyze high-throughput and well-standardized experimental studies in order to improve our understanding of sleep.Author summary: Machine learning-based approaches hold great promise to pave the way for high-throughput animal sleep monitoring. With the novel developments of gene-engineering techniques and the proliferation of experimental sleep studies, the need for the automation and cross-lab standardization of sleep scoring becomes more imminent. Traditionally, the classification of electroencephalographic (EEG) signatures is done by trained human experts via visual inspection. Here we present a novel algorithm based upon neural networks to automatically generate accurate and physiologically plausible predictions. Performed experiments demonstrate that the proposed solution offers de facto human level performance, is more accurate than any other approach to date (93-99% accurate compared to multiple trained human scorers), and functions across different genetic mutants, surgery procedures, recording setups and even different species. Moreover, our method was capable of detecting mutation-induced changes in sleeping patterns. To allow for its widespread adaptation, we make our framework freely available through the provision of an online server and an easy to use interface. This community tool will both contribute to the standardization of experimental studies and enhance scientific understanding of sleep.