Biology subjects

Yakovenko, S.

Publications and source records attributed to Yakovenko, S..

14 recordsLinked to original sources

bioRxiv2026

A phase decoder for reconstructing missing muscle activity

Upper-limb assistive devices are commonly controlled with electromyographic (EMG) signals, but nerve injury, muscle loss, or practical limits on electrode placement can leave only a few usable signals, which constrains the decoding of intended movements. Here, we introduce a decoder that reconstructs missing muscle activations from sparse EMGs using a two-stage real-time approach: (1) estimating ongoing movement phase from available signals, and (2) retrieving corresponding muscle activations from a library of pre-recorded muscle activity templates. In proof-of-concept tests, neurologically intact participants (N=4) performed object-transfer tasks under simulated denervation conditions. The decoder accurately estimated movement phase (median root-mean-square error [RMSE] <0.17 of movement duration) and reconstructed missing muscle activations (RMSE 0.08-0.34; correlation up to 0.97) within ~2 ms, in a previously unseen participant. The results indicate that template based reconstruction could leverage sparse EMG for low latency control of assistive technology.

neuroscience↗

bioRxiv2026

Neural networks learn forward dynamics when freed from numerical integration

Seamless interaction between humans and machines requires interfaces that remain robust to the variability inherent in biological signals and physical environments. Advanced human-machine interfaces (HMIs) increasingly rely on machine learning to predict or control limb dynamics. These systems must learn input-to-output mappings between control variables and limb state, such as the mapping from muscle forces or joint torques acting about segmented arm joints to limb posture over time. Such statistical input-to-output transformations can result in numerical instability of predicted musculoskeletal kinematics and dynamics. Achieving the robustness of biological motor control requires solving both forward and inverse dynamics problems; however, these problems are computationally asymmetric because they entail opposing operations-integration and differentiation. Since we have previously shown that neural networks solve the inverse dynamics problem when trained to map kinematic to dynamic signals during reaching, we hypothesized that representing separately the approximation of equations of motion (EOM) and their temporal numerical integration may capture the relevant computational structure of the forward dynamics problem. We tested this hypothesis by comparing a conventional direct-mapping recurrent neural network (RNN) with a two-stage model, the artificial physics engine (APE). When predicting the state of a two-segment system under external perturbations not encountered during training, the direct-mapping, monolithic model produced large prediction errors inconsistent with the expected interaction torque, whereas the APE maintained low error and remained stable under novel initial conditions and perturbations. Mapping system dynamics in the terms of the EOM improves robustness against intrinsic and extrinsic sources of variability by imposing a causal, physics-based structure on HMI design.

neuroscience↗

bioRxiv2024

Damage explains function in spiking neural networks representing central pattern generator

Complex biological systems evolved to control dynamics in the presence of noisy and often unpredictable inputs. The staple example is locomotor control, which is vital for survival. Control of locomotion results from interactions between multiple systems--from passive dynamics of inverted pendulum governing body motion to coupled neural oscillators that integrate predictive forward and sensory feedback signals. The neural dynamic computations are expressed in the rhythmogenic spinal network termed the central pattern generator (CPG). While a system of ordinary differential equations or a rate model is typically "good enough" to describe the CPG function, the computations performed by thousands of neurons in vertebrates are poorly understood. To study the distributed computations of a well-defined neural dynamic system, we developed a CPG model for gait expressed with the spiking neural networks (SNN). The SNN-CPG model faithfully recreated the input-output relationship of the rate model, describing the modulation of gait phase characteristics. The degradation of distributed computation within elements of the SNN-CPG model was further studied with "lesion" experiments. We found that lesioning flexor or extensor elements, with otherwise identical structural organization of reciprocal networks, affected differently the overall CPG computation. This result mimics experimental observations. Moreover, the increasing general excitability within the network can compensate for the loss of function after progressive lesions. This observation may explain the response to spinal stimulation and propose a novel theoretical framework for degraded computations and their applications within restorative technologies.

neuroscience↗

bioRxiv2024

Evidence of sensory error threshold in triggering locomotor adaptations in humans

Changing body biomechanics or external conditions trigger neural adaptations to optimize motor behavior. While the adaptations appear continuous to minimize movement errors, not all errors necessarily initiate sensorimotor adaptations. The locomotor system exhibits resilience to change and rehabilitation. This suggests the presence of an error threshold to trigger the adaptation mechanism. Here, we imposed kinematic and kinetic constraints of stepping using a passive orthosis and real-time feedback about limb loading. We aimed to manipulate stepping asymmetry and explore the interplay between adaptation and locomotor error magnitude. Uninjured healthy adults were tested in three locomotor conditions: unconstrained, constrained, and washout unconstrained walking. Surprisingly, kinematic asymmetry alone did not induce persistent adaptation. However, the addition of asymmetric interlimb loading triggered the expected adaptation. Our finding suggests that uninjured locomotor systems can cope with a specific range of kinematic asymmetries without initiating persistent adaptations that lead to aftereffects, and that loading may be a key variable for evoking the adaptation. The presence of an error threshold may mitigate possible disruption of vital motor functions and contribute to locomotor adaptation during walking. These insights elucidate the mechanism of neural plasticity and have implications for rehabilitation.

neuroscience↗

bioRxiv2023

Does joint impedance improve dynamic leg simulations with explicit and implicit solvers?

The nervous system predicts and executes complex motion of body segments actuated by the coordinated action of muscles. When a stroke or other traumatic injury disrupts neural processing, the impeded behavior has not only kinematic but also kinetic attributes that require interpretation. Biomechanical models could allow medical specialists to observe these dynamic variables and instantaneously diagnose mobility issues that may otherwise remain unnoticed. However, the real-time and subject-specific dynamic computations necessitate the optimization these simulations. In this study, we explored the effects of intrinsic viscoelasticity, choice of numerical integration method, and decrease in sampling frequency on the accuracy and stability of the simulation. The bipedal model with 17 rotational degrees of freedom (DOF)--describing hip, knee, ankle, and standing foot contact--was instrumented with viscoelastic elements with a resting length in the middle of the DOF range of motion. The accumulation of numerical errors was evaluated in dynamic simulations using swing-phase experimental kinematics. The relationship between viscoelasticity, sampling rates, and the integrator type was evaluated. The optimal selection of these three factors resulted in an accurate reconstruction of joint kinematics (err < 1%) and kinetics (err < 5%) with increased simulation time steps. Notably, joint viscoelasticity reduced the integration errors of explicit methods and had minimal to no additional benefit for implicit methods. Gained insights have the potential to improve diagnostic tools and accurize real-time feedback simulations used in the functional recovery of neuromuscular diseases and intuitive control of modern prosthetic solutions.

bioinformatics↗

bioRxiv2023

Artificial physics engine for real-time inverse dynamics of arm and hand movement

Simulating human body dynamics requires detailed and accurate mathematical models. When solved inversely, these models provide a comprehensive description of force generation that evaluates subject morphology and can be applied to control real-time assistive technology, for example, orthosis or muscle/nerve stimulation. Yet, model complexity hinders the speed of its computations and may require approximations as a mitigation strategy. Here, we use machine learning algorithms to provide a method for accurate physics simulations and subject-specific parameterization. Several types of artificial neural networks (ANNs) with varied architecture were tasked to generate the inverse dynamic transformation of realistic arm and hand movement (23 degrees of freedom). Using a physical model to generate the training and testing sets for the limb workspace, we developed the ANN transformations with low torque errors (less than 0.1 Nm). Multiple ANN implementations using kinematic sequences solved accurately and robustly the high-dimensional kinematic Jacobian and inverse dynamics of arm and hand. These results provide further support for the use of ANN architectures that use temporal trajectories of time-delayed values to make accurate predictions of limb dynamics.

neuroscience↗

bioRxiv2023

Intralimb locomotor coordination in rats walking on asymmetric pegway

Complex movements such as walking or reaching are generated by a sequence of muscle actions. How these coordinated actions subserve complex movements and their recovery after disruption remains unknown. The use of high throughput recording-stimulation systems with microelectrode access to structures along the neuraxis may complement the neurological models in rodents. To this purpose, we have trained rats to perform the precise foot placement locomotor task that allows us to assess skilled locomotor movements. Animals were pretrained on the peg walkway task, which was configured to impose either symmetric or asymmetric (with overstepping) locomotor stepping at preferred stride length. Selected forelimb muscles were implanted with intramuscular differential electrodes. After a week of recovery, we collected electromyography from the implanted muscles and ground reaction forces from the array of force sensors embedded into walkway pegs. The temporal relationship between muscle bursts was measured for each intralimb set of muscles (n=13) in symmetric and asymmetric stepping. The sequence corresponded to the progression of muscle actions responsible for limb lift, flexion and transport, overground clearance, and preparation for ground contact. The stereotyped spatiotemporal sequence of muscle activity was persistent and mirroring across the asymmetric tasks. These patterns are similar to those observed in cats during locomotion with and without obstacles and reaching movements. These findings support the hypothesis that the profiles of muscle activations are qualitatively similar across quadrupeds during precise locomotor tasks. New and NoteworthyWe characterize for the first time the spatiotemporal muscle activation in rat forelimb during precise asymmetric stepping on asymmetrically placed rungs. Similar to cats, the intralimb pattern of muscle activation in rats was stereotypical. The elements of this pattern were changing in a lateralized fashion based on the direction of the imposed asymmetry. The similarity of pattern to that of cats supports the idea of similarity of neural control across cat and rat species.

neuroscience↗

bioRxiv2022

Neuromechanical Coupling is Reflected in the Spatial Organization of the Spinal Motoneuron Pools

Recent advances in neuroprosthetics have shown that activating the surviving spinal circuitry below the level of spinal damage can produce complex movements and restore function. This is because the spinal cord circuitry plays an important role in the control of complex movements by integrating descending execution-related signals with feedback sensory signals. To generate efficient movements through muscle contractions, the nervous system embeds the anatomical and dynamical properties of the controlled body parts, a concept termed neuromechanical tuning. The motoneurons innervating the skeletal muscles of the body form complex spatial structures that span multiple segments. The goal of this study is to determine if these structures can support neuromechanical tuning. We performed a comparative analysis of the spatial organization of the motoneuron pools and the functional organization of the muscles they innervate. We developed a three-dimensional model of the spatial organization of motoneuron pools innervating upper limb muscles in the macaque and quantified their relative distances. We then utilized a musculoskeletal model of the macaque upper extremity to characterize the functional relationship between the muscles innervated by the motoneuron pools. We found that the distances between motoneuron pools mimicked the agonistic or antagonistic actions of muscles they innervate. This shows that the spatial organization of the motoneuron pools embeds the functional musculoskeletal anatomy evolved to support the large repertoire of upper limb movements in primates. This spinal anatomy may play an important role in the neuromechanical tuning and simplify the sensorimotor control of the arm.

neuroscience↗

bioRxiv2020

Solving musculoskeletal biomechanics with machine learning

Deep learning is a relatively new computational technique for the description of the musculoskeletal dynamics. The experimental relationships of muscle geometry in different postures are the high-dimensional spatial transformations that can be approximated by relatively simple functions, which opens the opportunity for machine learning applications. In this study, we challenged general machine learning algorithms with the problem of approximating the posture-dependent moment arm and muscle length relationships of the human arm and hand muscles. We used two types of algorithms, light gradient boosting machine (LGB) and fully connected artificial neural network (ANN) solving the wrapping kinematics of 33 muscles spanning up to six degrees of freedom (DOF) each for the arm and hand model with 18 DOFs. The input-output training and testing datasets were generated by our previous phenomenological model based on the autogenerated polynomial structures (Sobinov et al., 2019). Both models achieved a similar level of errors: ANN model errors were 0.08{+/-}0.05% for muscle lengths and 0.53{+/-}0.29% for moment arms, and LGB model made similar errors--0.18{+/-}0.06% and 0.13{+/-}0.07%, respectively. LGB model reached the training goal with only 10^3 samples, while ANN required 10^6 samples; however, LGB models were about 39 slower than ANN models in the evaluation. The sufficient performance of developed models demonstrates the future applicability of machine learning for musculoskeletal transformations in a variety of applications, such as in advanced powered prosthetics. Author SummaryThe accurate decoding of arm and hand motor intent from biological signals remains a key challenge. Solving this task with machine learning requires vast posture- and task-dependent data for identifying structural and functional parameters within dynamic musculoskeletal relationships. This problem is related to the curse of dimensionality where the processing complexity grows exponentially with the number of degrees of freedom described by the model. Here, we developed a tool based on artificial neural networks (ANN) to solve the kinematic transformation from posture to muscle path length and muscle moment arms. We used an accurate model of posture-dependent muscle moment arms and length to train and test the ability of ANN to solve this high-dimensional and computationally intense transformation and compare it to the boosted decision tree approach. We demonstrated that model-driven training is an efficient method to handle the encoding of high-dimensional musculoskeletal relationships. Adding muscles to the transformation, which increases the input-output complexity, does not reduce the prediction accuracy and does not require the increase in the number of elements within the network demonstrating the viability of this approach for applications using musculoskeletal biomechanics.

neuroscience↗

bioRxiv2020

Functional and Structural Moment Arm Validation for Musculoskeletal Models: A Study of the Human Forearm and Hand

The development of realistic musculoskeletal models is a fundamental goal for the theoretical progress in sensorimotor control and its engineering applications, e.g., in the biomimetic control of artificial limbs. Yet, accurate models require extensive experimental measures to validate structural and functional parameters describing muscle state over the full physiological range of motion. However, available experimental measurements of, for example, muscle moment arms are sparse and often disparate. Validation of these models is not trivial because of the highly complex anatomy and behavior of human limbs. In this study, we developed a method to validate and scale kinematic muscle parameters using posture-dependent moment arm profiles, isometric force measurements, and a computational detection of assembly errors. We used a previously published model with 18 degrees of freedom (DOFs) and 32 musculotendon actuators with force generated from a Hill-type muscle model. The muscle path from origin to insertion with wrapping geometry was used to model the muscle lengths and moment arms. We simulated moment arm profiles across the full physiological range of motion and compared them to an assembled dataset of published and merged experimental profiles. The muscle paths were adjusted using custom metrics based on root-mean-square and correlation coefficient of the difference between simulated and experimental values. Since the available measurements were sparse and the examination of high-dimensional muscles is challenging, we developed analyses to identify common failures, i.e., moment arm functional flipping due to the sign reversal in simulated moments and the imbalance of force generation between antagonistic groups in postural extremes. The validated model was used to evaluate the expected errors in torque generation with the assumption of constant instead of the posture-dependent moment arms at the wrist flexion-extension DOF, which is the critical joint in our model with the largest number of crossing muscles. We found that there was a reduction of joint torques by about 35% in the extreme quartiles of the wrist DOF. The use of realistic musculoskeletal models is essential for the reconstruction of hand dynamics. These models may improve the understanding of muscle actions and help in the design and control of artificial limbs in future applications. New & NoteworthyRealistic models of human limbs are a development goal required for the understanding of motor control and its applications in biomedical fields. However, developing accurate models is restrained by the lack of accurate and reliable musculoskeletal measurements in humans. Here, we have overcome this challenge by using multi-stage validation of simulated structures using both experimental data and the identification of structural failures in the high-dimensional muscle paths. We demonstrate that the rigorous structural and functional validation method is essential for the understanding of force generation at the wrist.

neuroscience↗

bioRxiv2019

Approximating complex musculoskeletal biomechanics using multidimensional autogenerating polynomials

Computational models of the musculoskeletal system are scientific tools used to study human movement, quantify the effects of injury and disease, and plan surgical interventions. Additionally, these models could also be used to intuitively link biological control signals and realistic high-dimensional articulated prosthetic limbs. However, implementing fast and accurate musculoskeletal computations that can be used to control a prosthetic limb in real-time is a challenging problem. As muscles typically span multiple joints, the wrapping over complex geometrical constraints changes their moment arms and length as a function of joint angle and, thus, their ability to generate joint torques. As a result of these biomechanical complexities, calculating these muscle state variables in real-time is a difficult simulation problem. Here, we report a method to accurately and efficiently calculate these variables for the forearm muscles that actuate the hand and wrist across multiple postures. The posture dependent muscle geometry, moment arms and lengths of modeled muscles, were captured using autogenerating polynomials that expanded their optimal selection of terms using information measurements. The iterative process approximated 33 musculotendon actuators, each spanning up to 6 DOFs in an 18 DOF model of the human arm and hand, defined over the full physiological range of motion. Using these polynomials, the entire forearm anatomy could be computed in <10 {micro}s, which is far better than what is required for real-time performance, and with low errors in moment arms (below 5%) and lengths (below 0.4%). Moreover, we demonstrate that the number of elements in these autogenerating polynomials does not increase exponentially with the increase in complexity of muscles, increasing linearly instead. The similar structure and function of muscles are represented with specific invariant polynomial terms. Dimensionality reduction using the polynomial terms alone resulted in clusters comprised of muscles with similar functions, suggesting that the polynomials themselves captured biologically relevant features of muscle structure and function. We propose that this novel method of describing musculoskeletal biomechanics might further improve the applications of detailed and scalable models for the description of human movement.

physiology↗

bioRxiv2018

Study of the DiI method accuracy on the example of the afferent connections of the Habenula in P6 F1-C57Bl6 / CBA mice

ABSTACTDiI (1,1-dioctadecyl-3,3,33-tetramethylindocarbocyanine perchlorate) is a lipophilic carbocyanine stain diffusing in the plasma membrane and coloring the cell as a whole.\n\nThe DiI injection method is an alternative to stereotaxis in the study of the neural connections of higher vertebrates. However, despite a number of studies performed to identify the neural connections of embryos and fetuses of different animals, the question of the accuracy of the DiI method remains open. The main goal of this study is to investigate the accuracy of the DiI method by examining the afferent Habenula (Hb) connections in P6 F1-C57Bl6/CBA mice. Hb is a paired structure of the intermediate brain, consisting of two nuclei - the medial (MHb) and lateral (LHb). At P6, the main Hb tracts are already formed but not yet myelinated. To estimate the accuracy of the method, two groups of cases with different types of application on different Hb nuclei were compared: bilateral applications of several DiI crystals - group 1 (26 applications); Unilateral applications of a single crystal DiI - control group 2 (7 applications). It is known that MHb is strictly innervated by Triangular septal nuclei (TS) and Septofimbrial septal nuclei (SFi); LHb is strictly innervated by the Lateral preopteic area (LPA). Any deviation from this scheme is a false positive result. When applying the marker to different Hb nuclei in group 1, the number of successful cases without deviations was -0%; in group 2 (control) - 28.57%. Unilateral applications of a single DiI crystal are more accurate. However, there are still a number of unavoidable problems, such as the fragmentation or loss of a single DiI crystal during application and the super-strong diffusion of the marker at the application site. Given this, the overall accuracy of the DiI injection method, even in the form of a single crystal, remains low.

neuroscience↗

bioRxiv2017

Analytical solution of a bilateral Brown-type central pattern generator for symmetric and asymmetric locomotion

The coordinated activity of muscles is produced in part by spinal rhythmogenic neural circuits, termed central pattern generators (CPGs). A classical CPG model, proposed conceptually by T.G. Brown in 1911, is a system of coupled oscillators that transform locomotor drive into coordinated and gait-specific patterns of muscle recruitment. A system of ordinary differential equations with a physiologically-inspired coupling locus of interactions captures the timing relationship for bilateral coordination of limbs in locomotion and is typically solved numerically. Consequently, it is intriguing to have a full analytical description of this plausible CPG architecture to illuminate the functionality within this structure. Here, we provided a closed-form analytical solution contrasted against the previous numerical method. The computational load of the analytical solution was decreased by an order of magnitude when compared to the numerical approach (relative errors, <0.01%). The analytical solution tested and supported the previous finding that the input to the model can be expressed in units of the desired limb locomotor speed. Furthermore, we performed parametric sensitivity analysis in the context of controlling asymmetric locomotion and documented two possible mechanisms associated with either an external drive or intrinsic CPG parameters. The results support the idea that many different combinations of network states, even within the same anatomical CPG structure, may generate the same behavioral outcomes.\n\nAuthor SummaryUsing a simple process of leaky integration, we developed an analytical solution to a robust model of spinal pattern generation. We analyzed the ability of this neural element to exert locomotor control of the signal associated with limb speeds, which represent high-level modality within the motor system. Furthermore, we tested the ability of this simple structure to embed steering control using the velocity signal in the models inputs or within the internal connectivity of its elements. Both mechanisms can produce the same behavioral outcome, pointing to the methodological challenges of modeling central pattern generators and demonstrating the possibility of spinal circuit adaptations to asymmetric short- or long-term conditions in health and disease.

neuroscience↗

bioRxiv2017

The Nonlinear Relationship Between Sensory And Motor Primitives During Reaching Movements

The stabilizing role of sensory feedback in relation to realistic 3-dimensional movement dynamics remains poorly understood. The objective of this study was to quantify how primary afferent activity contributes to shaping muscle activity patterns during reaching movements. To achieve this objective, we designed a virtual reality task that guided healthy human subjects through a set of planar reaching movements with controlled kinematic and dynamic conditions that minimized inter-subject variability. Next, we integrated human upper-limb models of musculoskeletal dynamics and proprioception to analyze motion and major muscle activation patterns during these tasks. We recorded electromyographic and motion-capture data and used the integrated model to simulate joint kinematics, joint torques due to muscle contractions, muscle length changes, and simulated primary afferent feedback. The parameters of the primary afferent model were altered systematically to evaluate the effect of fusimotor drive. The experimental and simulated data were analyzed with hierarchical clustering. We found that the muscle activity patterns contained flexible task-dependent groups that consisted of co-activating agonistic and antagonistic muscles that changed with the dynamics of the task. The activity of muscles spanning only the shoulder generally grouped into a proximal cluster, while the muscles spanning the wrist grouped into a distal cluster. The bifunctional muscle spanning the shoulder and elbow were flexibly grouped with either proximal or distal cluster based on the dynamical requirements of the task. The composition and activation of these groups reflected the relative contribution of active and passive forces to each motion. In contrast, the simulated primary afferent feedback was most related to joint kinematics rather than dynamics, even though the primary afferent models had nonlinear dynamical components and variable fusimotor drive. Simulated physiological changes to the fusimotor drive were not sufficient to reproduce the dynamical features in muscle activity pattern. Altogether, these results suggest that sensory feedback signals are in a different domain from that of muscle activation signals. This indicates that to solve the neuromechanical problem, the central nervous system controls limb dynamics through task-dependent co-activation of muscles and non-linear modulation of monosynaptic primary afferent feedback. New & NoteworthyHere we answered the fundamental question in sensorimotor transformation of how primary afferent signals can contribute to the compensation for limb dynamics evident in muscle activity. We combined computational and experimental approaches to create a new experimental paradigm that challenges the nervous system with passive limb dynamics that either assists or resists the desired movement. We found that the active dynamical features present in muscle activity are unlikely to arise from direct feedback from primary afferents.

neuroscience↗