OBJECTIVES:To determine if motor evoked potentials (MEP) - stimulation-induced muscle activation measured using electromyography - can serve as a biomarker of corticobulbar (CBT) and corticospinal (CST) tract activation induced by deep brain stimulation (DBS). METHODS:In 12 patients with Parkinson's disease and subthalamic (n = 10) or pallidal (n = 2) DBS, contact mapping determined clinical motor side effect thresholds. For equivalent stimulation parameters (except lower frequency), EMG was recorded from cranial and arm muscles to determine the presence, peak amplitudes and latencies of MEP. Clinical side effect and MEP thresholds were compared and accuracy metrics calculated. RESULTS:The MEP amplitudes increased with stimulation intensity. Latencies were shorter for cranial muscles, which were also more likely to generate an MEP. Clinical side effect and MEP thresholds were significantly correlated (R2 = 0.31; p = 0.0006), although most MEP thresholds were lower than side effect thresholds. The MEP accuracy in predicting side effects was 0.72, with a sensitivity of 0.68 and a specificity of 0.73. CONCLUSIONS:MEP thresholds correlated with side effect thresholds but were often lower indicating subclinical CBT/CST activations. SIGNIFICANCE:This study characterizes motor potentials evoked by DBS and demonstrates their potential utility as an objective biomarker for motor side effect detection during DBS programming.

Deep brain stimulation (DBS) is an effective treatment for Parkinson's disease (PD); however, there is limited understanding of which subthalamic pathways are recruited in response to stimulation. Here, by focusing on the polarity of the stimulus waveform (cathodic vs. anodic), our goal was to elucidate biophysical mechanisms that underlie electrical stimulation in the human brain. In clinical studies, cathodic stimulation more easily triggers behavioral responses, but anodic DBS broadens the therapeutic window. This suggests that neural pathways involved respond preferentially depending on stimulus polarity. To experimentally compare the activation of therapeutically relevant pathways during cathodic and anodic subthalamic nucleus (STN) DBS, pathway activation was quantified by measuring evoked potentials resulting from antidromic or orthodromic activation in 15 PD patients undergoing DBS implantation. Cortical evoked potentials (cEP) were recorded using subdural electrocorticography, DBS local evoked potentials (DLEP) were recorded from non-stimulating contacts and EMG activity was recorded from arm and face muscles. We measured: 1) the amplitude of short-latency cEP, previously demonstrated to reflect activation of the cortico-STN hyperdirect pathway, 2) DLEP amplitude thought to reflect activation of STN-globus pallidus (GP) pathway, and 3) amplitudes of very short-latency cEP and motor evoked potentials (mEP) for activation of cortico-spinal/bulbar tract (CSBT). We constructed recruitment and strength-duration curves for each EP/pathway to compare the excitability for different stimulation polarities. We compared experimental data with the most advanced DBS computational models. Our results provide experimental evidence that subcortical cathodic and anodic stimulation activate the same pathways in the STN region and that cathodic stimulation is in general more efficient. However, relative efficiency varies for different pathways so that anodic stimulation is the least efficient in activating CSBT, more efficient in activating the HDP and as efficient as cathodic in activating STN-GP pathway. Our experiments confirm biophysical model predictions regarding neural activations in the central nervous system and provide evidence that stimulus polarity has differential effects on passing axons, terminal synapses, and local neurons. Comparison of experimental results with clinical DBS studies provides further evidence that the hyperdirect pathway may be involved in the therapeutic mechanisms of DBS.

During cortical spreading depolarization (CSD), neurons exhibit a dramatic increase in cytosolic calcium, which may be integral to CSD-mediated seizure termination. This calcium increase greatly exceeds that during seizures, suggesting the calcium source may not be solely extracellular. Thus, we sought to determine if the endoplasmic reticulum (ER), the largest intracellular calcium store, is involved. We developed a two-photon calcium imaging paradigm to simultaneously record the cytosol and ER during seizures in awake mice. Paired with direct current recording, we reveal that CSD can manifest as a slow post-ictal cytosolic calcium wave with a concomitant depletion of ER calcium that is spatiotemporally consistent with a calcium-induced calcium release. Importantly, we observed both naturally occurring and electrically induced CSD suppressed post-ictal epileptiform activity. Collectively, this work links ER dynamics to CSD, which serves as an innate process for seizure suppression and a potential mechanism underlying therapeutic electrical stimulation for epilepsy.

Modulating brain oscillations has strong therapeutic potential. Interventions that both non-invasively modulate deep brain structures and are practical for chronic daily home use are desirable for a variety of therapeutic applications. Repetitive audio-visual stimulation, or sensory flicker, is an accessible approach that modulates hippocampus in mice, but its effects in humans are poorly defined. We therefore quantified the neurophysiological effects of flicker with high spatiotemporal resolution in patients with focal epilepsy who underwent intracranial seizure monitoring. In this interventional trial (NCT04188834) with a cross-over design, subjects underwent different frequencies of flicker stimulation in the same recording session with the effect of sensory flicker exposure on local field potential (LFP) power and interictal epileptiform discharges (IEDs) as primary and secondary outcomes, respectively. Flicker focally modulated local field potentials in expected canonical sensory cortices but also in the medial temporal lobe and prefrontal cortex, likely via resonance of stimulated long-range circuits. Moreover, flicker decreased interictal epileptiform discharges, a pathological biomarker of epilepsy and degenerative diseases, most strongly in regions where potentials were flicker-modulated, especially the visual cortex and medial temporal lobe. This trial met the scientific goal and is now closed. Our findings reveal how multi-sensory stimulation may modulate cortical structures to mitigate pathological activity in humans. Repetitive audio-visual stimulation, or sensory flicker, can modulate oscillations in a non-invasive manner. Here the authors demonstrate the potential of flicker in individuals with epilepsy undergoing intracranial seizure monitoring.

Objective.To treat neurological and psychiatric diseases with deep brain stimulation (DBS), a trained clinician must select parameters for each patient by monitoring their symptoms and side-effects in a months-long trial-and-error process, delaying optimal clinical outcomes. Bayesian optimization has been proposed as an efficient method to quickly and automatically search for optimal parameters. However, conventional Bayesian optimization does not account for patient safety and could trigger unwanted or dangerous side-effects.Approach.In this study we develop SAFE-OPT, a Bayesian optimization algorithm designed to learn subject-specific safety constraints to avoid potentially harmful stimulation settings during optimization. We prototype and validate SAFE-OPT using a rodent multielectrode stimulation paradigm which causes subject-specific performance deficits in a spatial memory task. We first use data from an initial cohort of subjects to build a simulation where we design the best SAFE-OPT configuration for safe and accurate searchingin silico. Main results.We then deploy both SAFE-OPT and conventional Bayesian optimization without safety constraints in new subjectsin vivo, showing that SAFE-OPT can find an optimally high stimulation amplitude that does not harm task performance with comparable sample efficiency to Bayesian optimization and without selecting amplitude values that exceed the subject's safety threshold.Significance.The incorporation of safety constraints will provide a key step for adopting Bayesian optimization in real-world applications of DBS.

Temporal Interference (TI) is an emerging method to non-invasively stimulate deep brain structures. This innovative technique is increasingly recognized for its potential applications in the treatment of various neurological disorders, including epilepsy, depression, and Alzheimer’s disease. However, several drawbacks to the TI method exist that we aim to improve upon. To begin, the applied electric field in the TI target is not much higher than what non-invasive transcranial alternating current stimulation (TACS) provides in the cortex. Additionally, the TI stimulation onset is dependent on the envelope of the amplitude modulated (AM) signal, where for example 1 Hz and 100 Hz envelopes have significantly different rise times to reach maximum envelope amplitude – unlike square biphasic pulses. This limitation in turn prevents classic TI, from applying bursts of pulses. Finally, the electric field intensity of TI cannot be increased or decreased at the target without dramatically altering the spatial profile of the stimulation focus. In the work presented here, we efficiently address all three of these limitations. First, we performed two-photon calcium imaging to show that individual neurons selectively respond to the TI envelope frequency, providing evidence that TI modulates neural activity with temporal specificity. This marks a significant advancement, representing the first empirical demonstration of neuronal activation at the Δf frequency within the context of TI and in an imaging modality. Subsequently, we compared the AM signals of TI with phase-shift keying (PSK) modulated signals to highlight the superior effectiveness of noninvasive pulses in contrast to the traditional TI method, particularly in inducing epileptic activity (after-discharges) in mice. We also added a multipolar configuration to create a significant increase in the electric field at the target without significantly altering the spatial profile and applied Fourier components to replicate classic biphasic bursts of square pulses - all transcranially, without the use of penetrating electrodes. These innovations aim to enhance the precision and efficacy of TI stimulation, to advance its application in neurological research and therapy.Key Points / Highlights 1. Non-invasive temporal interference stimulation modulates the activity of individual neurons at the envelope frequency.2. A non-invasive multi-pulse TI stimulation paradigm improves both temporal and spatial focality in the deep target neural tissue when compared to traditional continuous wave (amplitude-modulated) TI stimulation.3. Pulse TI paradigms can stimulate deep neural targets with reduced amplitude of the topical high-frequency stimulation, decreasing off-target stimulation when compared to continuous wave TI patterns. As a consequence, pulse TI stimulation reduces the risk of undesired side effects such as high-frequency conduction block in off-target tissues or cortical areas.4. Both temporal and spatial focality of the TI stimulation pattern positively correlate with the efficacy of the stimulation to induce seizures in the mouse hippocampus.### Competing Interest StatementEN has a minority stake in TI Solutions, which manufactures TI hardware to support TI research.

Objective. Therapeutic brain stimulation is conventionally delivered using constant-frequency stimulation pulses. Several recent clinical studies have explored how unconventional and irregular temporal stimulation patterns could enable better therapy. However, it is challenging to understand which irregular patterns are most effective for different therapeutic applications given the massively high-dimensional parameter space.Approach. Here we applied many irregular stimulation patterns in a single neural circuit to demonstrate how they can enable new dimensions of neural control compared to conventional stimulation, to guide future exploration of novel stimulation patterns in translational settings. We optogenetically excited the septohippocampal circuit with constant-frequency, nested pulse, sinusoidal, and randomized stimulation waveforms, systematically varying their amplitude and frequency parameters.Main results.We first found equal entrainment of hippocampal oscillations: all waveforms provided similar gamma-power increase, whereas no parameters increased theta-band power above baseline (despite the mechanistic role of the medial septum in driving hippocampal theta oscillations). We then compared each of the effects of each waveform on high-dimensional multi-band activity states using dimensionality reduction methods. Strikingly, we found that conventional stimulation drove predominantly 'artificial' (different from behavioral activity) effects, whereas all irregular waveforms induced activity patterns that more closely resembled behavioral activity.Significance. Our findings suggest that irregular stimulation patterns are not useful when the desired mechanism is to suppress or enhance a single frequency band. However, novel stimulation patterns may provide the greatest benefit for neural control applications where entraining a particular mixture of bands (e.g. if they are associated with different symptoms) or behaviorally-relevant activity is desired.

Brain stimulation holds promise for treating brain disorders, but personalizing therapy remains challenging. Effective treatment requires establishing a functional link between stimulation parameters and brain response, yet traditional methods like random sampling (RS) are inefficient and costly. To overcome this, we developed an active learning (AL) framework that identifies optimal relationships between stimulation parameters and brain response with fewer experiments. We validated this framework through three experiments: (1) in silico modeling with synthetic data from a Parkinson’s disease model, (2) in silico modeling with real data from a non-human primate, and (3) in vivo modeling with a real-time rat optogenetic stimulation experiment. In each experiment, we compared AL models to RS models, using various query strategies and stimulation parameters (amplitude, frequency, pulse width). AL models consistently outperformed RS models, achieving lower error on unseen test data in silico (p<0.0056, N=1000) and in vivo (p=0.0036, N=20). This approach represents a significant advancement in brain stimulation, potentially improving both research and clinical applications by making them more efficient and effective. Our findings suggest that AL can substantially reduce the cost and time required for developing personalized brain stimulation therapies, paving the way for more effective and accessible treatments for brain disorders.

Neural responses to pulses of electrical stimulation, termed "evoked potentials", can map brain connectivity and optimize deep brain stimulation as used in the treatment of neurological disease. As human neurotechnology now allows for simultaneous real-time sensing and stimulation at multiple channels throughout the brain, it will benefit from automated real-time detection of evoked potentials to prospectively guide brain stimulation targeting. Here we used intracranial brain stimulation data collected from 22 epilepsy patients undergoing seizure monitoring to design and evaluate an automated strategy for detecting evoked potentials produced by electrical brain stimulation. We evaluate and demonstrate the utility of two features - a high-frequency broadband power ratio, and cross-correlation across repeated stimulation trials - in detecting evoked potentials, showing that cross-correlation is a robust feature that can achieve 93% detection accuracy alone. We also show that combining these complementary features into a single metric improves detection performance over single features, and we present a complementary strategy for stimulation artifact rejection that improves detection performance of all features. In conclusion, we present an automated strategy for detecting evoked potentials that can be applied to large-scale brain data and used online to optimize brain stimulation targeting in applications such as Parkinson's disease, epilepsy, and more.

Significance:Intravital cellular calcium imaging has emerged as a powerful tool to investigate how different types of neurons interact at the microcircuit level to produce seizure activity, with newfound potential to understand epilepsy. Although many methods exist to measure seizure-related activity in traditional electrophysiology, few yet exist for calcium imaging. Aim:To demonstrate an automated algorithmic framework to detect seizure-related events using calcium imaging-including the detection of pre-ictal spike events, propagation of the seizure wavefront, and terminal spreading waves for both population-level activity and that of individual cells. Approach:We developed an algorithm for precise recruitment detection of population and individual cells during seizure-associated events, which broadly leverages averaged population activity and high-magnitude slope features to detect single-cell pre-ictal spike and seizure recruitment. We applied this method to data recorded using awake in vivo two-photon calcium imaging during pentylenetetrazol-induced seizures in mice. Results:We demonstrate that our detected recruitment times are concordant with visually identified labels provided by an expert reviewer and are sufficiently accurate to model the spatiotemporal progression of seizure-associated traveling waves. Conclusions:Our algorithm enables accurate cell recruitment detection and will serve as a useful tool for researchers investigating seizure dynamics using calcium imaging.

Brain stimulation has become an important treatment option for a variety of neurological and psychiatric diseases. A key challenge in improving brain stimulation is selecting the optimal set of stimulation parameters for each patient, as parameter spaces are too large for brute-force search and their induced effects can exhibit complex subject-specific behavior. To achieve greatest effectiveness, stimulation parameters may additionally need to be adjusted based on an underlying neural state, which may be unknown, unmeasurable, or challenging to quantify a priori. In this study, we first develop a simulation of a state-dependent brain stimulation experiment using rodent optogenetic stimulation data. We then use this simulation to demonstrate and evaluate two implementations of an adaptive Bayesian optimization algorithm that can model a dynamically changing response to stimulation parameters without requiring knowledge of the underlying neural state. We show that, while standard Bayesian optimization converges and overfits to a single optimal set of stimulation parameters, adaptive Bayesian optimization can continue to update and explore as the neural state is changing and can provide more accurate optimal parameter estimation when the optimal stimulation parameters shift. These results suggest that learning algorithms such as adaptive Bayesian optimization can successfully find optimal state-dependent stimulation parameters, even when brain sensing and decoding technologies are insufficient to track the relevant neural state.

Electrical brain stimulation has become an essential treatment option for more than one third of epilepsy patients who are resistant to pharmacological therapy and are not candidates for surgical resection. However, currently approved stimulation paradigms achieve only moderate success, on average providing approximately 75% reduction in seizure frequency and extended periods of seizure freedom in nearly 20% of patients. Outcomes from electrical stimulation may be improved through the identification of novel anatomical targets, particularly those with significant anatomical and functional connectivity to the epileptogenic zone. Multiple studies have investigated the medial septal nucleus (i.e., medial septum) as such a target for the treatment of mesial temporal lobe epilepsy. The medial septum is a small midline nucleus that provides a critical functional role in modulating the hippocampal theta rhythm, a 4–7‐Hz electrophysiological oscillation mechanistically associated with memory and higher order cognition in both rodents and humans. Elevated theta oscillations are thought to represent a seizure‐resistant network activity state, suggesting that electrical neuromodulation of the medial septum and restoration of theta‐rhythmic physiology may not only reduce seizure frequency, but also restore cognitive comorbidities associated with mesial temporal lobe epilepsy. Here, we review the anatomical and physiological function of the septohippocampal network, evidence for seizure‐resistant effects of the theta rhythm, and the results of stimulation experiments across both rodent and human studies, to argue that deep brain stimulation of the medial septum holds potential to provide an effective neuromodulation treatment for mesial temporal lobe epilepsy. We conclude by discussing the considerations necessary for further evaluating this treatment paradigm with a clinical trial.

Introduction Brain stimulation is a fundamental and effective therapy for neurological diseases including Parkinson’s disease, essential tremor, and epilepsy. One key challenge in delivering effective brain stimulation is identifying the stimulation parameters, such as the amplitude, frequency, contact configuration, and pulse width, that induce an optimal change in symptoms, behavior, or neural activity. Most clinical and translational studies use constant-frequency pulses of stimulation, but stimulation with irregular pulse patterns or non-pulsatile waveforms might induce unique changes in neural activity that could enable better therapeutic responses. Here, we comprehensively evaluate several optogenetic stimulation waveforms, report their differing effects on hippocampal spectral activity, and compare these induced effects to activity recorded during natural behavior. Methods Sprague-Dawley rats were prepared for pan-neuronal excitatory optogenetic stimulation of the medial septum (hSyn-ChR2) and 16-channel microelectrode recording in CA1 and CA3 layers of the hippocampus. We performed grid and random sampling of the parameters comprising several stimulation waveforms, including standard pulse, nested pulse, sinusoid, double sinusoid, and Poisson pulse waveforms. Results We comprehensively report the effects of changing stimulation parameters in these parameter spaces on two key biomarkers of hippocampal function, theta (4-10 Hz) and gamma (32-50 Hz) power. Similarly, robust excitation of hippocampal gamma power was observed across all waveforms, whereas no set of stimulation parameters was sufficient to consistently increase power in the theta band beyond baseline levels of activity (despite the prominent role of the medial septum in pacing hippocampal theta oscillations). Using a manifold learning algorithm to compare high-dimensional neural activity, we show that irregular stimulation patterns produce differing effects with respect to multi-band patterns of activity and can induce activity patterns that more closely resemble activity recorded during natural behavior than conventional parameters. Conclusion Our counter-intuitive findings – that stimulation of the medial septum ubiquitously does not increase hippocampal theta power, and that different waveforms have similar effects on single power bands – contradict recent trends in brain stimulation research, necessitating greater caution and fewer mechanistic assumptions as to how a given stimulation target or waveform will modulate a neurophysiological biomarker of disease. We also reveal that irregular stimulation patterns can have biomimetic utility, promoting their exploration in medical applications where inducing a particular activity pattern can have therapeutic benefit. Last, we demonstrate a scalable data-driven analysis strategy that can make the discovery of such physiologically informed temporal stimulation patterns more empirically tractable in translational settings. ### Competing Interest Statement The authors have declared no competing interest.

Neural modulation is a fundamental tool for treating neurological diseases and understanding their mechanisms. One of the challenges in neural modulation includes selecting stimulation parameters, as parameter spaces are very large and their induced effects can exhibit complex behavior. Moreover, the effect of stimulation may depend on the underlying neural state, which can be difficult or impossible to quantify a priori . In this study, we first use an unsupervised learning approach to demonstrate that the effect of medial septum optogenetic stimulation on hippocampal activity differs between awake and anesthetized behavioral states. We then use these data to construct a simulation model of a neural modulation experiment and demonstrate a novel Bayesian optimization method that automatically learns the subject-specific relationship between neural state and its effect on modulation. This approach outperformed standard Bayesian optimization and identified ground-truth optimal parameters of the simulation model, suggesting that this method can efficiently explore complex state-dependent relationships of parameter spaces to improve neural modulation.

Neural modulation has become a fundamental tool for treating neurological diseases and understanding their mechanisms. A significant challenge in neural modulation is understanding and characterizing the brain's response to stimulation parameters - a difficult problem when parameter spaces are very large, their evaluation is time-consuming, and their effects feature complex inter- and intra- subject variability. In this study, we first fit a model to optogenetic stimulation data to construct a simulation of a neural modulation experiment. We then show two ways that Gaussian process modeling can supplement neural modulation data collection procedures - by quantifying the uncertainty of standard sampling approaches and by selecting samples via active learning. We found that the active learning approach required 33% fewer samples than the other strategies to accurately learn the mean of the simulation function. We also show that the uncertainty estimate of the active learning model converged most similarly to the ground-truth error throughout the simulation. These results suggest that Gaussian process modeling can quantitatively improve data collection procedures in neural modulation.

s Brain Stimulation 14 (2021) 1589e1707 BETH ISRAEL DEACONNESS MEDICAL CENTER, Milton, MA, USA; Yale School of Medicine, Boston, MA, USA; 6 Turku University Central Hospital Paimio Hospital, Bethesda, MD, Finland Abstract Background: Focal epilepsy is increasingly conceptualized as a disease of brain networks, but the network localization remains poorly understood. Lesion locations associated with epilepsy may help identify epilepsy networks and lead to new treatment targets. Methods: We studied 701 stroke patients and analyzed the lesion locations associated with epilepsy (n 1⁄4 76) or control (n 1⁄4 625). Lesion locations were mapped to a common brain atlas and the brain network functionally connected to each lesion locationwas computed using human connectome data (n 1⁄4 1000). Functional connections associated with stroke-related epilepsy were identified using lesion-network mapping. Generalizability to epilepsy associated with other lesion types (n 1⁄4 452) was assessed using a leave-one-lesion-type-out cross-validation. Finally, therapeutic relevance of these connections was assessed using outcome data from patients who received thalamic deep brain stimulation for drug resistant epilepsy (n 1⁄4 30). Results: Lesion locations associated with stroke-related epilepsy map to a specific brain network defined by functional connectivity to nodes in the cerebellum, substantia nigra, and globus pallidus (P < 0.01). Connectivity to this network was associated with the risk of epilepsy across different lesion types (P 1⁄4 0.0001) and with therapeutic response to thalamic deep brain stimulation (R 1⁄4 0.63, P 1⁄4 0.00017). Conclusions: Brain lesions associated with epilepsy map to a specific brain network, with therapeutic potential for neuromodulation in epilepsy.

When a robotic system is faced with uncertainty, the system must take calculated risks to gain information as efficiently as possible while ensuring system safety. The need to safely and efficiently gain information in the face of uncertainty spans domains from healthcare to search and rescue. To efficiently learn when data is scarce or difficult to label, active learning acquisition functions intelligently select a data point that, if the label were known, would most improve the estimate of the unknown model. Unfortunately, prior work in active learning suffers from an inability to accurately quantify information-gain, generalize to new domains, and ensure safe operation. To overcome these limitations, we develop Safe MetAL, a probabilistically-safe, active learning algorithm which meta-learns an acquisition function for selecting sample efficient data points in safety critical domains. The key to our approach is a novel integration of meta-active learning and chance-constrained optimization. We (1) meta-learn an acquisition function based on sample history, (2) encode this acquisition function in a chance-constrained optimization framework, and (3) solve for an information-rich set of data points while enforcing probabilistic safety guarantees. We present state-of-the-art results in active learning of the model of a damaged UAV and in learning the optimal parameters for deep brain stimulation. Our approach achieves a 41% improvement in learning the optimal model and a 20% speedup in computation time compared to active and meta-learning approaches while ensuring safety of the system.

Objective. Deep brain stimulation (DBS) is an effective treatment for Parkinson’s disease (PD) but its success depends on a time-consuming process of trial-and-error to identify the optimal stimulation settings for each individual patient. Data-driven optimization algorithms have been proposed to efficiently find the stimulation setting that maximizes a quantitative biomarker of symptom relief. However, these algorithms cannot efficiently take into account stimulation settings that may control symptoms but also cause side effects. Here we demonstrate how multi-objective data-driven optimization can be used to find the optimal trade-off between maximizing symptom relief and minimizing side effects. Approach. Cortical and motor evoked potential data collected from PD patients during intraoperative stimulation of the subthalamic nucleus were used to construct a framework for designing and prototyping data-driven multi-objective optimization algorithms. Using this framework, we explored how these techniques can be applied clinically, and characterized the design features critical for solving this optimization problem. Our two optimization objectives were to maximize cortical evoked potentials, a putative biomarker of therapeutic benefit, and to minimize motor potentials, a biomarker of motor side effects. Main Results. Using this in silico design framework, we demonstrated how the optimal trade-off between two objectives can substantially reduce the stimulation parameter space by 61 ± 19%. The best algorithm for identifying the optimal trade-off between the two objectives was a Bayesian optimization approach with an area under the receiver operating characteristic curve of up to 0.94 ± 0.02, which was possible with the use of a surrogate model and a well-tuned acquisition function to efficiently select which stimulation settings to sample. Significance. These findings show that multi-objective optimization is a promising approach for identifying the optimal trade-off between symptom relief and side effects in DBS. Moreover, these approaches can be readily extended to newly discovered biomarkers, adapted to DBS for disorders beyond PD, and can scale with the development of more complex DBS devices.