Rory John Bufacchi

Introduction: Adverse social experiences profoundly alter animal behavior, yet the processes underlying context-appropriate behavior selection based on prior social interactions remain poorly understood. Existing models capture statistical patterns but are not framed in mechanistic frameworks that explain individual variability and predict behavioral outcomes. Methods: We address this gap by modeling social defeat in mice as a partially observable Markov decision process (POMDP), implementing a heterarchical agent architecture—a structured network of interacting modules balancing exploration and exploitation. Our model successfully reconstructs observed behavioral motifs (e.g., investigation, hesitation, and flights), fits different mouse phenotypes (e.g., susceptible vs. resilient), and mechanistically captures the impact of social defeat as a parameter shift in the animal’s internal generative model. Notably, characteristic behavioral motifs including investigation bouts, hesitation, and flight trajectories emerge from the model’s inference dynamics, without any trajectory-level fitting. The model additionally reproduces effects of interventions like optogenetic stimulation, and generates testable predictions for future experiments. The model’s modular architecture enables natural extension to other behavioral domains including foraging and multi-agent interactions, representing a foundational step toward interpretable models of mouse behavior. By capturing how adverse social experiences reshape decision-making at the computational level, this work offers potential clinical relevance for trauma and anxiety disorders in humans.

1. Adverse social experiences profoundly alter animal behavior, yet the processes underlying context-appropriate behavior selection based on prior social interactions remain poorly understood. Existing models capture statistical patterns but are not framed in mechanistic frameworks that explain individual variability and predict behavioral outcomes. We address this gap by modeling social defeat in mice as a partially observable Markov decision process (POMDP), implementing a heterarchical agent architecture - a structured network of interacting modules balancing exploration and exploitation. Our model successfully reconstructs observed behavioral motifs (e.g., investigation, hesitation, and flights), fits different mouse phenotypes (e.g., susceptible vs. resilient), and mechanistically captures the impact of social defeat as a parameter shift in the animal’s internal generative model. The model reproduces effects of interventions like optogenetic stimulation, and generates testable predictions for future experiments. The model’s modular architecture enables natural extension to other behavioral domains including foraging and multi-agent interactions, representing a foundational step toward interpretable models of mouse behavior. By capturing how adverse social experiences reshape decision-making at the computational level, this work offers potential clinical relevance for trauma and anxiety disorders in humans.

The modulation of the hand‐blink reflex (HBR), a prototypical nocifensive response, is increasingly used to investigate defensive behaviour, related to the notion of peripersonal space. However, HBR responsiveness is highly variable across participants. This variability has led researchers to use several seemingly arbitrary criteria to determine whether a subject should be included as a participant in a study. But are these criteria justified? Can better and more rigorous criteria be formulated? Does the traditional division into responders and non‐responders reflect a practical decision to exclude participants with very low signal‐to‐noise ratio, or does it reflect two distinct biological categories? Here, we addressed these issues by systematically varying a set of parameters, which together form an objective and quantifiable criterion of HBR responsiveness. We describe classification criteria for HBR responsiveness that were both reliable and consistent with previous studies. We also found no evidence for a clear‐cut biological distinction between HBR responders and non‐responders. We recommend to (1) no longer preliminarily screen subjects, simply collecting data on all subjects, and (2), after collecting the data, only include subjects identified as blinkers using the following criteria: The mean of the rectified HBR must exceed 2.5 SD of the baseline EMG in 40% or more of trials in the hand‐near condition. We formulate rigorous inclusion criteria for HBR studies, which can be adapted for use on other neurophysiological responses in health and disease.

Body-part-centered response fields are pervasive in single neurons, functional magnetic resonance imaging, electroencephalography and behavior, but there is no unifying formal explanation of their origins and role. In the present study, we used reinforcement learning and artificial neural networks to demonstrate that body-part-centered fields do not simply reflect stimulus configuration, but rather action value: they naturally arise from the basic assumption that agents often experience positive or negative reward after contacting environmental objects. This perspective successfully reproduces experimental findings that are foundational in the peripersonal space literature. It also suggests that peripersonal fields provide building blocks that create a modular model of the world near the agent: an egocentric value map. This concept is strongly supported by the emergent modularity that we observed in our artificial networks. The short-term, close-range, egocentric map is analogous to the long-term, long-range, allocentric hippocampal map. This perspective fits empirical data from multiple experiments, provides testable predictions and accommodates existing explanations of peripersonal fields.

Body-part centric response fields are pervasive: they are observed in single neurons, fMRI, EEG, and multiple behavioural measures. This prevalence across scales and measures makes them excellent candidates for studying systems-level neuroscience. Nonetheless, they remain poorly understood because we lack a unifying formal explanation of their origins and role in wider brain function. Here, we provide such explanation. We use reinforcement learning to analytically explain the existence of body-part centric receptive fields, also known as peripersonal field. We then simulate multiple experimental findings considered foundational in the peripersonal space literature. Our results demonstrate that peripersonal fields naturally arise from two simple and plausible assumptions about living agents: 1) they experience reward when they contact objects in the environment, and 2) they act to maximise reward. These simple assumptions are enough to explain empirical findings on stimulus kinematics, tool use, valence, and network-architecture. Our explanation provides further insight. First, it offers multiple empirically testable predictions. Second, it offers a formal description of the notion that the world-agent state is encoded in parieto-premotor cortices, using motor primitives: peripersonal fields provide building blocks that together create a short-term model of the world near the agent in terms of its future states; a successor representation. This short-term, close-range egocentric peripersonal map is analogous to the long-term, long-range allocentric spatial map of place and grid cells, which underlie locomotion and navigation to reach distant objects. Together, these allocentric and egocentric maps allow efficient interactions with a changing environment across multiple spatial and temporal scales. Abstract Figure:

Abrupt increases of sensory input (onsets) likely reflect the occurrence of novel events or objects in the environment, potentially requiring immediate behavioral responses. Accordingly, onsets elicit a transient and widespread modulation of ongoing electrocortical activity: the Vertex Potential (VP), which is likely related to the optimisation of rapid behavioral responses. In contrast, the functional significance of the brain response elicited by abrupt decreases of sensory input (offsets) is more elusive, and a detailed comparison of onset and offset VPs is lacking. In four experiments conducted on 44 humans, we observed that onset and offset VPs share several phenomenological and functional properties: they (1) have highly similar scalp topographies across time, (2) are both largely comprised of supramodal neural activity, (3) are both highly sensitive to surprise and (4) co-occur with similar modulations of ongoing motor output. These results demonstrate that the onset and offset VPs largely reflect the activity of a common supramodal brain network, likely consequent to the activation of the extralemniscal sensory system which runs in parallel with core sensory pathways. The transient activation of this system has clear implications in optimizing the behavioral responses to surprising environmental changes.

Shadmehr and Ahmed cogently argue that vigor of appetitive movements is positively correlated with their value, and that value can therefore be inferred by measuring vigor. Here, we highlight three points to consider when interpreting this account: (1) The correlation between vigor and value is not obligatory, (2) the vigor effect also arises in frameworks other than optimal foraging, and (3) the term vigor can be misinterpreted, thereby affecting rigor.

Living in rapidly changing environments has shaped the mammalian brain toward high sensitivity to abrupt and intense sensory events—often signaling threats or affordances requiring swift reactions. Unsurprisingly, such events elicit a widespread electrocortical response (the vertex potential, VP), likely related to the preparation of appropriate behavioral reactions. Although the VP magnitude is largely determined by stimulus intensity, the relative contribution of the differential and absolute components of intensity remains unknown. Here, we dissociated the effects of these two components. We systematically varied the size of abrupt intensity increases embedded within continuous stimulation at different absolute intensities, while recording brain activity in humans (with scalp electroencephalography) and rats (with epidural electrocorticography). We obtained three main results. 1) VP magnitude largely depends on differential, and not absolute, stimulus intensity. This result held true, 2) for both auditory and somatosensory stimuli, indicating that sensitivity to differential intensity is supramodal, and 3) in both humans and rats, suggesting that sensitivity to abrupt intensity differentials is phylogenetically well-conserved. Altogether, the current results show that these large electrocortical responses are most sensitive to the detection of sensory changes that more likely signal the sudden appearance of novel objects or events in the environment.

Nervous systems exploit regularities in the sensory environment to predict sensory input and adjust behavior, and thereby maximize fitness. Entrainment of neural oscillations allows retaining temporal regularities of sensory information, a prerequisite for prediction. Entrainment has been extensively described at the frequencies of periodic inputs most commonly present in visual and auditory landscapes (e.g. >1 Hz). An open question is whether neural entrainment also occurs for regularities at much longer timescales. Here we exploited the fact that the temporal dynamics of thermal stimuli in natural environment can unfold very slowly. We show that ultra-low frequency neural oscillations preserved a long-lasting trace of sensory information through neural entrainment to periodic thermo-nociceptive input as low as 0.1 Hz. Importantly, revealing the functional significance of this phenomenon, both power and phase of the entrainment predicted individual pain sensitivity. In contrast, periodic auditory input at the same ultra-low frequency did not entrain ultra-low frequency oscillations. These results demonstrate that a functionally-significant neural entrainment can occur at temporal scales far longer than those commonly explored. The non-supramodal nature of our results suggests that ultra-low frequency entrainment might be tuned to the temporal scale of the statistical regularities characteristic of different sensory modalities.