Novel and optimized mouse behavior enabled by fully autonomous HABITS (Home-cage assisted behavioral innovation and testing system)

Authors: Bowen Yu, Penghai Li, Haoze Xu, Yueming Wang, Kedi Xu, Yaoyao Hao (Zhejiang University; Nanhu Brain-computer Interface Institute). Journal: eLife 14:RP104833 (2025 Sep 16). DOI 10.7554/eLife.104833. PMCID PMC12440354 (open access, CC BY). Preprint: bioRxiv 10.1101/2024.09.29.615652.

Abstract

Mice are among the most prevalent animal models used in neuroscience, benefiting from the extensive physiological, imaging, and genetic tools available to study their brain. However, the development of novel and optimized behavioral paradigms for mice has been laborious and inconsistent, impeding the investigation of complex cognitions. Here, we present a home-cage assisted mouse behavioral innovation and testing system (HABITS), enabling free-moving mice to learn challenging cognitive behaviors in their home-cage without any human involvement. Supported by the general programming framework, we have not only replicated established paradigms in current neuroscience research but also developed novel paradigms previously unexplored in mice, resulting in more than 300 mice demonstrated in various cognition functions. Most significantly, HABITS incorporates a machine-teaching algorithm, which comprehensively optimized the presentation of stimuli and modalities for trials, leading to more efficient training and higher-quality behavioral outcomes. To our knowledge, this is the first instance where mouse behavior has been systematically optimized by an algorithmic approach. Altogether, our results open a new avenue for mouse behavioral innovation and optimization, which directly facilitates investigation of neural circuits for novel cognitions with mice.

eLife digest

Mice are widely used in neuroscience research due to the many tools available to study their brain function and behavior. However, training mice for complex tasks requires extensive human involvement, which can stress the animals and introduce inconsistencies in methods and results.

Automated systems can reduce any potential bias, but most focus on single tasks only and lack optimization. To address these issues, Yu et al. developed the Home-cage Assisted Behavioral Innovation and Testing System (HABITS) – a fully autonomous platform where mice learn tasks in their cages without human intervention.

Using HABITS, mice successfully acquired a wide range of cognitive skills – including decision-making, working memory, and attention – entirely without handling. The system uses machine learning to adjust training sequences, improving learning speed and minimizing bias. In tests with over 300 mice across more than 20 paradigms, including some never attempted in mice, HABITS also improves the overall health of mice compared to the conventionally used water-restriction training. The system’s AI-driven adjustments help mice learn challenging tasks more efficiently and with fewer errors.

Its low cost and automation make it an efficient and reliable tool to study behavior, making it suitable for large-scale studies. Future developments may incorporate wireless neural recordings to directly link behavior with brain activity, providing deeper insight into learning and decision-making mechanisms.

Key points for the wiki

• Fully autonomous home-cage training: free-moving mice learn challenging cognitive tasks in their home cage with no human involvement; 300+ mice tested across 20+ paradigms (including paradigms never previously attempted in mice).
• Machine-teaching algorithm optimises stimulus presentation/modality for trials — reported as the first systematic algorithmic optimisation of mouse behaviour training; faster training, higher-quality data.
• Improved welfare compared with conventional water-restriction training.
• Open source: github.com/Yaoyao-Hao/HABITS; general state-machine runner gpSMART on GitHub; data on figshare.