Acquiring musculoskeletal skills with curriculum-based reinforcement learning.

Efficient musculoskeletal simulators and powerful learning algorithms provide computational tools to tackle the grand challenge of understanding biological motor control. Our winning solution for the inaugural NeurIPS MyoChallenge leverages an approach mirroring human skill learning. Using a novel curriculum learning approach, we trained a recurrent neural network to control a realistic model of the human hand with 39 muscles to rotate two Baoding balls in the palm of the hand. In agreement with data from human subjects, the policy uncovers a small number of kinematic synergies, even though it is not explicitly biased toward low-dimensional solutions. However, selectively inactivating parts of the control signal, we found that more dimensions contribute to the task performance than suggested by traditional synergy analysis. Overall, our work illustrates the emerging possibilities at the interface of musculoskeletal physics engines, reinforcement learning, and neuroscience to advance our understanding of biological motor control.

Decoding the brain: From neural representations to mechanistic models.

A central principle in neuroscience is that neurons within the brain act in concert to produce perception, cognition, and adaptive behavior. Neurons are organized into specialized brain areas, dedicated to different functions to varying extents, and their function relies on distributed circuits to continuously encode relevant environmental and body-state features, enabling other areas to decode (interpret) these representations for computing meaningful decisions and executing precise movements. Thus, the distributed brain can be thought of as a series of computations that act to encode and decode information. In this perspective, we detail important concepts of neural encoding and decoding and highlight the mathematical tools used to measure them, including deep learning methods. We provide case studies where decoding concepts enable foundational and translational science in motor, visual, and language processing.

Keypoint-MoSeq: parsing behavior by linking point tracking to pose dynamics.

Keypoint tracking algorithms can flexibly quantify animal movement from videos obtained in a wide variety of settings. However, it remains unclear how to parse continuous keypoint data into discrete actions. This challenge is particularly acute because keypoint data are susceptible to high-frequency jitter that clustering algorithms can mistake for transitions between actions. Here we present keypoint-MoSeq, a machine learning-based platform for identifying behavioral modules ('syllables') from keypoint data without human supervision. Keypoint-MoSeq uses a generative model to distinguish keypoint noise from behavior, enabling it to identify syllables whose boundaries correspond to natural sub-second discontinuities in pose dynamics. Keypoint-MoSeq outperforms commonly used alternative clustering methods at identifying these transitions, at capturing correlations between neural activity and behavior and at classifying either solitary or social behaviors in accordance with human annotations. Keypoint-MoSeq also works in multiple species and generalizes beyond the syllable timescale, identifying fast sniff-aligned movements in mice and a spectrum of oscillatory behaviors in fruit flies. Keypoint-MoSeq, therefore, renders accessible the modular structure of behavior through standard video recordings.

SuperAnimal pretrained pose estimation models for behavioral analysis.

Quantification of behavior is critical in diverse applications from neuroscience, veterinary medicine to animal conservation. A common key step for behavioral analysis is first extracting relevant keypoints on animals, known as pose estimation. However, reliable inference of poses currently requires domain knowledge and manual labeling effort to build supervised models. We present SuperAnimal, a method to develop unified foundation models that can be used on over 45 species, without additional manual labels. These models show excellent performance across six pose estimation benchmarks. We demonstrate how to fine-tune the models (if needed) on differently labeled data and provide tooling for unsupervised video adaptation to boost performance and decrease jitter across frames. If fine-tuned, SuperAnimal models are 10-100× more data efficient than prior transfer-learning-based approaches. We illustrate the utility of our models in behavioral classification and kinematic analysis. Collectively, we present a data-efficient solution for animal pose estimation.

Altered thermal preferences of infected or immune-challenged Aedes aegypti and Aedes japonicus mosquitoes.

Temperature is a critical factor shaping physiology, life cycle, and behaviour of ectothermic vector insects, as well as the development and multiplication of pathogens within them. However, the influence of pathogen infections on thermal preferences (behavioural thermoregulation) is not well-understood. The present study examined the thermal preferences of mosquitoes (Aedes aegypti and Ae. japonicus) infected with either Sindbis virus (SINV) or Dirofilaria immitis over 12 days post exposure (p.e.) or injected with a non-pathogenic Sephadex bead over 24 h in a thermal gradient (15-30 °C). SINV-infected Ae. aegypti preferred 5 °C warmer temperatures than non-infected ones at day 6 p.e., probably the time of highest innate immune response. In contrast, D. immitis-infected Ae. japonicus preferred 4 °C cooler temperatures than non-infected ones at day 9 p.e., presumably a stress response during the migration of third instar larvae from their development site to the proboscis. Sephadex bead injection also induced a cold preference in the mosquitoes but to a level that did not differ from control-injections. The cold preference thus might be a strategy to escape the risk of desiccation caused by the wound created by piercing the thorax. Further research is needed to uncover the genetic and physiological mechanisms underlying these behaviours.

Task-driven neural network models predict neural dynamics of proprioception.

Proprioception tells the brain the state of the body based on distributed sensory neurons. Yet, the principles that govern proprioceptive processing are poorly understood. Here, we employ a task-driven modeling approach to investigate the neural code of proprioceptive neurons in cuneate nucleus (CN) and somatosensory cortex area 2 (S1). We simulated muscle spindle signals through musculoskeletal modeling and generated a large-scale movement repertoire to train neural networks based on 16 hypotheses, each representing different computational goals. We found that the emerging, task-optimized internal representations generalize from synthetic data to predict neural dynamics in CN and S1 of primates. Computational tasks that aim to predict the limb position and velocity were the best at predicting the neural activity in both areas. Since task optimization develops representations that better predict neural activity during active than passive movements, we postulate that neural activity in the CN and S1 is top-down modulated during goal-directed movements.

Thermal preference of Culicoides biting midges in laboratory and semi-field settings.

Biting midges of the genus Culicoides (Diptera: Ceratopogonidae) are hematophagous insects, and some species can transmit a plethora of pathogens, e.g., bluetongue virus and African horse sickness virus, that mainly affect animals. The transmission of vector-borne pathogens is strongly temperature dependent, and recent studies pointed to the importance of including microclimatic data when modelling disease spread. However, little is known about the preferred temperature of biting midges. The present study addressed the thermal selection of field-caught Culicoides with two experiments. In a laboratory setup, sugar-fed or blood-fed midges were video tracked for 15 min while moving inside a 60 × 30 × 4 cm setup with a 15-25 °C temperature gradient. Culicoides spent over double the time in the coldest zone of the setup compared to the warmest one. This cold selection was significantly stronger for sugar-fed individuals. Calculated preferred temperatures were 18.3 °C and 18.9 °C for sugar-fed and blood-fed Culicoides, respectively. The effect of temperature on walking speed was significant but weak, indicating that their skewed distribution results from preference and not cold trapping. A second experiment consisted of a two-way-choice-setup, performed in a 90 × 45 × 45 cm net cage, placed outdoors in a sheltered environment. Two UV LED CDC traps were placed inside the setup, and a mean temperature difference of 2.2 °C was created between the two traps. Hundred-fifty Culicoides were released per experiment. Recapture rates were negatively correlated with ambient temperature and were on average three times higher in the cooled trap. The higher prevalence of biting midges in cooler environments influences fitness and ability to transmit pathogens and should be considered in models that predict Culicoides disease transmission.

The Phlebotomine sand fly fauna of Switzerland revisited.

Sand flies (Diptera: Psychodidae, Phlebotominae; Newstead, 1911) are widespread in Europe, being particularly common in the Mediterranean region but rare north of the Alps. Thus, Switzerland is an opportune place to investigate the sand fly fauna on both sides of the Alpine crest, in southern sub-Mediterranean climate and northern oceanic temperate climate. We reinvestigated the Swiss sand fly fauna with the aim to assess changes in composition, altitudinal distribution, abundance and seasonality. Thirty-eight sites were investigated with light traps and/or interception sticky traps in 4 years. Ninety and 380 specimens were caught by light traps and sticky traps, respectively, at 15 collecting sites. Four species were identified. Phlebotomus mascittii (Grassi, 1908), Phlebotomus perniciosus (Newstead, 1911) and Sergentomyia minuta (Rondani, 1843) were confirmed in Ticino, and P. mascittii for the first time in neighbouring Grisons. Also, Phlebotomus neglectus (Tonnoir, 1921) is for the first time reported, though at a very low density compared to P. perniciosus at the same site. Its presence in Ticino supports the northward spread observed in Italy. Sand flies were detected north of the Alps at one site only, endorsing a historical report. Overall, the low density of P. perniciosus and very low density of P. neglectus suggest that canine leishmaniosis may not be an important disease risk in Switzerland.

Environment rather than breed or body site shapes the skin bacterial community of healthy sheep as revealed by metabarcoding.

BACKGROUND: The skin is inhabited by a variety of micro-organisms, with bacteria representing the predominant taxon of the skin microbiome. In sheep, the skin bacterial community of healthy animals has been addressed in few studies, only with culture-based methods or sequencing of cloned amplicons. OBJECTIVES: The objectives of this study were to determine the sheep skin bacterial community composition by using metabarcoding for a detailed characterisation and to determine the effect of body part, breed and environment. MATERIALS AND METHODS: Overall, 267 samples were taken from 89 adult female sheep, belonging to three different breeds and kept on nine different farms in Switzerland. From every individual, one sample each was taken from belly, left ear and left leg and metabarcoding of the 16S rRNA V3-V4 hypervariable region was performed. RESULTS: The main phyla identified were Actinobacteriota, Firmicutes, Proteobacteria and Bacteriodota. The alpha diversity as determined by Shannon's diversity index was significantly different between sheep from different farms. Beta diversity analysis by principal coordinate analysis (PCoA) showed clustering of the samples by farm and body site, while breed had only a marginal influence. A sparse partial least squares discriminant analysis (sPLS-DA) revealed seven main groups of operational taxonomic units (OTUs) of which groups of OTUs were specific for some farms. CONCLUSIONS AND CLINICAL RELEVANCE: These findings indicate that environment has a larger influence on skin microbial variability than breed, although the sampled breeds, the most abundant ones in Switzerland, are phenotypically similar. Future studies on the sheep skin microbiome may lead to novel insights in skin diseases and prevention.

Temperature preference of sugar- or blood-fed Aedes japonicus mosquitoes under semi-natural conditions.

Mosquito-borne diseases pose a major burden on humans and animals. Temperature strongly influences the physiology and life cycle of mosquitoes and also the pathogens they transmit. Thermoregulatory behaviour of mosquitoes has been addressed in a few laboratory studies. Here, we expand such studies by investigating the thermal preference when resting of Aedes japonicus, an invasive and putative vector species of many pathogens, in a semi-field setup during summers in a temperate climate. Blood-fed or sugar-fed Ae. japonicus females were released in the late afternoon in a large outdoor cage containing three resting boxes. The next morning, temperature treatments were applied to the boxes, creating a "cool" (over all experiments around 18 °C), and a "warm" (around 35 °C) microhabitat in addition to an untreated "ambient" (around 26 °C) one. The mosquitoes resting within the three boxes were counted five times, every 2 h between 9h and 17h. The highest proportions of mosquitoes (e.g. up to 21% of blood-fed ones) were found in the cool box while both blood-fed and sugar-fed mosquitoes avoided the warm box. The mean resting temperatures of Ae. japonicus were below the ambient temperatures measured by a nearby meteorological station, and this was more pronounced at higher outdoor temperatures and in blood-fed as compared to sugar-fed mosquitoes. Thus, over all experiments with blood-fed mosquitoes, the calculated average resting temperature was 4 °C below the outdoor temperature. As mosquitoes prefer cooler resting places than temperatures measured by weather stations in summer, models to predict mosquito-borne disease outbreaks need to account for the thermoregulatory behaviour of mosquitoes, especially in the wake of climate change.

Contrasting action and posture coding with hierarchical deep neural network models of proprioception.

Biological motor control is versatile, efficient, and depends on proprioceptive feedback. Muscles are flexible and undergo continuous changes, requiring distributed adaptive control mechanisms that continuously account for the body's state. The canonical role of proprioception is representing the body state. We hypothesize that the proprioceptive system could also be critical for high-level tasks such as action recognition. To test this theory, we pursued a task-driven modeling approach, which allowed us to isolate the study of proprioception. We generated a large synthetic dataset of human arm trajectories tracing characters of the Latin alphabet in 3D space, together with muscle activities obtained from a musculoskeletal model and model-based muscle spindle activity. Next, we compared two classes of tasks: trajectory decoding and action recognition, which allowed us to train hierarchical models to decode either the position and velocity of the end-effector of one's posture or the character (action) identity from the spindle firing patterns. We found that artificial neural networks could robustly solve both tasks, and the networks' units show tuning properties similar to neurons in the primate somatosensory cortex and the brainstem. Remarkably, we found uniformly distributed directional selective units only with the action-recognition-trained models and not the trajectory-decoding-trained models. This suggests that proprioceptive encoding is additionally associated with higher-level functions such as action recognition and therefore provides new, experimentally testable hypotheses of how proprioception aids in adaptive motor control.

Sense the moist and feel the heat: mosquito host and oviposition site detection.

Laursen et al. identified coreceptor Ir93a that is required for humidity and thermal sensing in the mosquito species Anopheles gambiae and Aedes aegypti. Behavioural studies with mutant mosquitoes with disrupted Ir93a revealed that they were less attracted to a blood meal source and oviposition site at close distance.

Keypoint-MoSeq: parsing behavior by linking point tracking to pose dynamics.

Keypoint tracking algorithms have revolutionized the analysis of animal behavior, enabling investigators to flexibly quantify behavioral dynamics from conventional video recordings obtained in a wide variety of settings. However, it remains unclear how to parse continuous keypoint data into the modules out of which behavior is organized. This challenge is particularly acute because keypoint data is susceptible to high frequency jitter that clustering algorithms can mistake for transitions between behavioral modules. Here we present keypoint-MoSeq, a machine learning-based platform for identifying behavioral modules ("syllables") from keypoint data without human supervision. Keypoint-MoSeq uses a generative model to distinguish keypoint noise from behavior, enabling it to effectively identify syllables whose boundaries correspond to natural sub-second discontinuities inherent to mouse behavior. Keypoint-MoSeq outperforms commonly used alternative clustering methods at identifying these transitions, at capturing correlations between neural activity and behavior, and at classifying either solitary or social behaviors in accordance with human annotations. Keypoint-MoSeq therefore renders behavioral syllables and grammar accessible to the many researchers who use standard video to capture animal behavior.

NeuroAI: If grid cells are the answer, is path integration the question?

Spatially modulated neurons known as grid cells are thought to play an important role in spatial cognition. A new study has found that units with grid-cell-like properties can emerge within artificial neural networks trained to path integrate, and developed a unifying theory explaining the formation of these cells which shows what circuit constraints are necessary and how learned systems carry out path integration.

Morphological description and multilocus genotyping of Onchocerca spp. in red deer (Cervus elaphus) in Switzerland.

Onchocercosis is a parasitic disease caused by over 30 Onchocerca spp. (Nematoda: Filarioidea) and predominantly affecting ungulates. Four Onchocerca spp. have been described in the European red deer (Cervus elaphus). Onchocerca flexuosa and Onchocerca jakutensis form subcutaneous nodules in the back region. The other two species, Onchocerca skrjabini and the lesser-known Onchocerca garmsi, are found freely in the subcutaneous tissue of carpal and tarsal joints, and the sternal region, respectively. The presence of Onchocerca spp. in eight red deer shot in the hunting season during September 2020 in the Grisons region, Switzerland, was investigated by analysing nodules and free worms in the subcutaneous tissue. The obtained worms were morphologically and genetically identified as O. jakutensis, O. flexuosa and O. skrjabini. The latter two are first reports from Switzerland, and morphological redescriptions of these two species are presented. Onchocerca skrjabini and O. jakutensis are newly described from the sternal region of deer. One female of O. jakutensis was found free in the subcutaneous tissue of the sternal region, an atypical presentation for this species. Phylogenetic analyses were based on four mitochondrial and one nuclear loci, revealing that O. jakutensis belongs to a clade which so far only included non-cervid Onchocerca spp. Analysis of sequences from this study and GenBank entries revealed two distinct subpopulations of O. skrjabini: one from European red deer and another from Japanese serow and sika deer. Morphological identification can be challenging, also because worm location in the host is less strictly determined than previously described. Genetic identification is straightforward for O. flexuosa, O. jakutensis and O. skrjabini for which complete data of five loci are now available whereas genetic data of O. garmsi is still lacking.

Striatal dopamine explains novelty-induced behavioral dynamics and individual variability in threat prediction.

Animals both explore and avoid novel objects in the environment, but the neural mechanisms that underlie these behaviors and their dynamics remain uncharacterized. Here, we used multi-point tracking (DeepLabCut) and behavioral segmentation (MoSeq) to characterize the behavior of mice freely interacting with a novel object. Novelty elicits a characteristic sequence of behavior, starting with investigatory approach and culminating in object engagement or avoidance. Dopamine in the tail of the striatum (TS) suppresses engagement, and dopamine responses were predictive of individual variability in behavior. Behavioral dynamics and individual variability are explained by a reinforcement-learning (RL) model of threat prediction in which behavior arises from a novelty-induced initial threat prediction (akin to "shaping bonus") and a threat prediction that is learned through dopamine-mediated threat prediction errors. These results uncover an algorithmic similarity between reward- and threat-related dopamine sub-systems.

Putative roles of mosquitoes (Culicidae) and biting midges (Culicoides spp.) as mechanical or biological vectors of lumpy skin disease virus.

The stable fly Stomoxys calcitrans (Diptera: Muscidae) is considered as the main mechanical vector of the lumpy skin disease virus (LSDV). In addition, the mosquito species Aedes aegypti (Diptera: Culicidae) was shown to transmit the virus from donor to receptor animals. Retention of the virus for several days was shown for two additional tropical mosquito species and the biting midge Culicoides nubeculosus (Diptera: Ceratopogonidae). In the present study, viral retention for 10- or 7-days post feeding on virus-spiked blood through a membrane was shown for field-collected Aedes japonicus and laboratory-reared Culex pipiens, two widely distributed mosquito species in temperate regions. Viral DNA could be detected from honey-coated Flinders Technology Associates (FTA) cards and shedded faeces for 1 or 4 days after an infectious blood meal was given to Ae. aegypti. Virus increase over time and virus dissemination was observed in laboratory-reared C. nubeculosus, but the virus could be isolated from field-collected biting midges only from the day of exposure to the blood meal. Thus, mosquitoes might serve as mechanical vectors of LSDV in case of interrupted feeding. A putative biological virus transmission by Culicoides biting midges, as suspected from field observations, deserves further investigations.

Video analysis of the locomotory behaviour of Aedes aegypti and Ae. japonicus mosquitoes under different temperature regimes in a laboratory setting.

Mosquito-borne diseases impose a high burden on human and animal health. Temperature strongly influences the physiology and life cycle of mosquitoes, but also the development and/or propagation of the pathogens they transmit. Thus, the vector capacity of mosquitoes depends strongly on temperature and their behavioural thermoregulation through microhabitat selection. Expanding on a previous study of static thermal preferences, the locomotory dynamics of temperate Aedes japonicus (reared from eggs collected in the field) and tropical Ae. aegypti (from a laboratory colony) was investigated at constant temperatures (10 °C, 25 °C, 40 °C) and in temperature gradients (10-20 °C, 20-30 °C, 30-40 °C). Blood-fed or non-blood-fed female mosquitoes were released in groups of 15 individuals into a Plexiglas box positioned on two thermoregulators connected by an aluminium plate to automatically monitor by video analysis mosquito flying, walking and resting duration, covered distances and velocity. Mosquitoes were predominantly resting, followed by walking and flying. At constant 10 °C, flights were rare and brief, and walking was slow. Most activity was observed at 25 °C for Ae. japonicus and 40 °C for Ae. aegypti. In the 30-40 °C gradient, activity of Ae. aegypti increased towards the cold end, suggesting active avoidance of very high temperatures. In the 20-30 °C gradient, edge effects were prominent, nevertheless revealing a greater proportion of mosquitoes gathered at the cooler end. Video analysis showed that this effect was not caused by a cold trap but represents true thermal preference. In the coolest gradient (10-20 °C), mosquitoes were active in all sectors without displaying a preference for either side. Overall, both the tropical and temperate mosquito species preferred cooler temperatures and actively avoided the hottest temperatures. Further studies with infected mosquitoes should provide important insights for developing models of vector-borne disease outbreaks.

Multi-animal pose estimation, identification and tracking with DeepLabCut.

Estimating the pose of multiple animals is a challenging computer vision problem: frequent interactions cause occlusions and complicate the association of detected keypoints to the correct individuals, as well as having highly similar looking animals that interact more closely than in typical multi-human scenarios. To take up this challenge, we build on DeepLabCut, an open-source pose estimation toolbox, and provide high-performance animal assembly and tracking-features required for multi-animal scenarios. Furthermore, we integrate the ability to predict an animal's identity to assist tracking (in case of occlusions). We illustrate the power of this framework with four datasets varying in complexity, which we release to serve as a benchmark for future algorithm development.

Perspectives in machine learning for wildlife conservation.

Inexpensive and accessible sensors are accelerating data acquisition in animal ecology. These technologies hold great potential for large-scale ecological understanding, but are limited by current processing approaches which inefficiently distill data into relevant information. We argue that animal ecologists can capitalize on large datasets generated by modern sensors by combining machine learning approaches with domain knowledge. Incorporating machine learning into ecological workflows could improve inputs for ecological models and lead to integrated hybrid modeling tools. This approach will require close interdisciplinary collaboration to ensure the quality of novel approaches and train a new generation of data scientists in ecology and conservation.