Abstract Working out whether others care for us is crucial in personal relationships and when seeking professional help. It is often most difficult for those most in need, e.g. following interpersonal traumata. Here, we introduce a simple ‘Caring Attributions task’ and elucidate key computational mechanisms involved and, using EEG, the cortical activity representing the degree of belief that another is beneficent or maleficent. We find evidence for a new type of neurocomputational processing: valence-partitioned temporal-difference inference (TD-Bayes). This employs primary processing about latent causes, but also separate channels to represent these different valenced attributions, inspired by value-partitioned associative learning (VPAL). TD-Bayes uses slow propagation of beliefs using temporal-difference updating. These models gave a very good account of behaviour, slightly better than VPAL, but crucially, their partitioned representations have stronger, distinct representations in ERP signals. They provide a promising inroad into the understanding of how people may jump to atttibutions about caring vs. uncaring others.

Understanding others' intentions amidst uncertainty is critical for effective social interactions, yet the neural mechanisms underlying this process are not fully understood. Here, we combined computational modeling and single-trial EEG analysis to examine how the brain dynamically updates beliefs about others' intentions in volatile social contexts. A total of 43 healthy volunteers engaged in a deception-free advice-taking task, featuring alternating stable and volatile phases that systematically manipulated the reliability of an adviser's intentions. Using the hierarchical Gaussian filter (HGF), a Bayesian model of learning, we quantified trial-by-trial updates of participants' beliefs and their neural correlates. EEG amplitudes systematically varied according to task volatility, engaging neural regions associated with uncertainty processing such as the fusiform gyrus and posterior cingulate cortex. Sensor-level EEG analyses confirmed a temporal sequence consistent with the hierarchical computations predicted by the HGF, whereby lower-level prediction errors were processed earlier than higher-order volatility-related signals. Moreover, individual differences in these hierarchical neural processes correlated significantly with psychosocial functioning, suggesting that disruptions in Bayesian belief updating may underlie functional impairments in clinical populations. Collectively, our results reveal novel neural evidence for hierarchical Bayesian inference during social learning, highlighting its critical role in adaptive social behavior and potential relevance to mental health.

Anhedonia, a core symptom of mood disorders such as depression, is marked by diminished pleasure and motivation for rewards. Traditional reinforcement learning (RL) tasks, like the 4-arm bandit (4AB), show limited sensitivity to reward-processing impairments associated with anhedonia. Here, we developed and validated a modified 3-arm bandit (3AB) task designed to reduce cognitive load while retaining sensitivity to reward and punishment learning. Following this validation, 1,000 participants were pre-screened with the Snaith-Hamilton Pleasure Scale (SHAPS), Dimensional Anhedonia Rating Scale (DARS), Generalized Anxiety Disorder Assessment (GAD), and Zung Self-Rating Depression Scale (ZUNG). Participants scoring above 2 on SHAPS and below 45 on DARS were classified as anhedonic, while those scoring 0 on SHAPS and above 55 on DARS were classified as non-anhedonic, resulting in 111 anhedonic and 95 non-anhedonic individuals who completed the 3AB task. Modelling results revealed no significant group differences in reward learning rate (p = 0.23), punishment learning rate (p = 0.37), reward sensitivity (p = 0.28), or punishment sensitivity (p = 0.46). Additional assessments of win-stay/lose-shift strategies and reaction times also showed no significant differences. Bayes Factor t-tests provided moderate -to-strong evidence for the null hypothesis, with BF01 values of 3.36, 5.14, 4.97, and 5.96 for each of the above parameters, respectively. These findings indicate either that anhedonia does not impair reward and punishment learning or that the 3AB task lacks the sensitivity to detect such differences. By simplifying the task structure while maintaining core learning mechanisms, the 3AB task provides a novel approach for studying reward processing, challenging the notion that anhedonia impairs reward sensitivity and indicating that this learning mechanism may remain intact.

The ventral tegmental area is the primary source of dopaminergic input to the human prefrontal cortex and plays a central role in reinforcement learning. Although animal studies have established that dopaminergic neurons encode reward prediction error signals, direct electrophysiological evidence in humans is scarce. Understanding these mechanisms is clinically relevant because of their involvement in disorders of motivation and reward processing. In this cross-sectional study, we recorded local field potentials from the ventral tegmental area in fourteen patients (nine male; mean age 46 years, range 30–62) undergoing deep brain stimulation surgery for chronic cluster headache. During temporary electrode externalisation, participants performed a probabilistic instrumental learning task comprising reward, loss and neutral trials. Behaviour was modelled using a hierarchical Rescorla–Wagner framework with separate learning rates for rewards and losses. Electrophysiological responses were analysed using Statistical Parametric Mapping and linear mixed-effects models testing sensitivity to outcome, expected value and reward prediction error. Clear evoked responses were observed for stimulus and outcome events in thirteen subjects. Responses reflecting the contrast between outcomes that delivered a gain, a loss or a neutral signal and those delivering no outcome were significantly larger for gains than for losses or neutral signals (paired t-tests: gain versus loss, t(12)=2.60, p=0.023, Cohen’s d=0.72; gain versus neutral, t(12)=4.49, p<0.001, d=1.25), while loss and neutral contrasts did not differ (p=0.218). Nine of fourteen subjects developed a clear preference for the high-reward option; two exhibited gradual learning, while others adopted a win-stay strategy. In eight subjects with clear evoked responses, activity around the button press correlated with the expected value of the chosen option (peak at 0.02s, p=0.013, corrected). In a further subset of two subjects who explored the low-value option, single-trial responses peaked closer to the button press during low-value choices. No compelling evidence was found for a distinct reward prediction error signal beyond outcome and value. Clinical covariates and smoking status did not significantly modulate electrophysiological responses. Human ventral tegmental area activity is selectively tuned to rewarding outcomes and, under conditions of effective learning, reflects the expected value of chosen options during decision-making. These findings align with reinforcement learning principles established in animal models and suggest that local field potentials primarily represent inputs to reward prediction error computation rather than its output. This work provides the first direct electrophysiological evidence of reward-related signalling in the human ventral tegmental area, supporting its translational relevance for understanding motivation and guiding neuromodulatory interventions.

Anhedonia, a transdiagnostic symptom marked by diminished reward sensitivity, is often linked to impairments in reinforcement learning (RL). Standard tasks (e.g., the 4-arm bandit) can place substantial demands on participants and may blur valuation with other processes. We therefore adapted a three-arm bandit (3AB) task from Seymour et al. (2012), incorporating design features intended to lessen task demands (fewer options; denser feedback) while enabling separate estimation of reward and punishment learning rates and sensitivities. In an online sample pre-screened for anhedonia (N = 206; 111 anhedonic, 95 non-anhedonic), hierarchical Bayesian modelling using a four-parameter specification showed no credible group differences in reward learning rate, punishment learning rate, reward sensitivity, or punishment sensitivity; Bayes factors favoured the null (BF01 = 3.36–5.96). Model-agnostic win-stay/lose-shift strategies likewise showed no group differences (Welch’s tests, all p > .05). Posterior predictive checks indicated above-chance choice prediction: the model’s highest-probability action matched participants’ actual choices on 59.6% of trials (chance = 33%). Parameter recovery was excellent for valuation parameters (r = 0.96–0.97) and acceptable for learning rates (r = 0.67–0.85). Simulations generated from fitted parameters preserved individual-difference structure, with high correlations between observed and simulated win-stay (r = 0.89 anhedonic; 0.86 non-anhedonic) and moderate correlations for lose-shift (r = 0.62; 0.67), alongside small systematic mean-level biases (simulated win-stay lower by 3.5–4.9 percentage points; simulated lose-shift higher by 12.8–13.2 points). Model comparison showed that lapse-augmented variants achieved marginally better predictive fit, but group comparisons under both lapse models yielded overlapping posteriors with 95% HDIs including zero for all learning, sensitivity, and lapse parameters, indicating that the null findings were robust to inclusion of lapse terms. Non-anhedonic participants also responded more slowly on average than anhedonic participants, which we treat as exploratory. Together, these results suggest that in this 3AB task, anhedonia is not reliably associated with differences in core RL parameters or simple choice strategies, while providing a detailed characterisation of model performance and limitations in an online setting.

BACKGROUND:The corticobasal ganglia network in Parkinson's disease (PD) is characterized by the occurrence of transient episodes of exaggerated beta frequency oscillatory synchrony, known as bursts. Although it is known that bursts of prolonged duration associate closely with motor impairments, the mechanisms leading to burst initiation remain poorly understood. Related to this, current adaptive deep brain stimulation (DBS) approaches reactively deliver stimulation following burst onset but cannot stimulate proactively to prevent bursts from occurring. The discovery of predictive biomarkers could allow for proactive stimulation, thereby offering potential for improvements in therapeutic efficacy. OBJECTIVES:We aimed to address this issue, by using deep neural networks to discover features of basal ganglia activity that reliably precede beta burst onset. METHODS:We developed a deep learning model to predict burst onset from subthalamic nucleus (STN) activity recordings in PD patients. Our model provides patient-specific predictions in two independent datasets of STN recordings, including prolonged-duration recordings from sensing-enabled DBS devices during naturalistic behaviors. RESULTS:The occurrence of STN beta bursts can be reliably predicted up to 100 ms prior to onset. Importantly, our results reveal that a dip in the beta amplitude-which is likely to be indicative of a phase reset of oscillatory populations occurring between 80 and 100 ms prior to burst onset-is a predictive biomarker for burst occurrence. CONCLUSIONS:These findings demonstrate proof-of-principle for the feasibility of beta burst prediction and inform the future development of intelligent DBS approaches with the capability of proactive stimulation to prevent beta burst occurrence. © 2025 The Author(s). Movement Disorders published by Wiley Periodicals LLC on behalf of International Parkinson and Movement Disorder Society.

This paper marks the 30th anniversary of the Statistical Parametric Mapping (SPM) software and the journal Cerebral Cortex: two modest milestones that mark the inception of cognitive neuroscience. We take this opportunity to reflect on SPM, a generation after its introduction. Each of the authors of this paper-who represent a small selection of the many contributors to SPM-were asked to consider lessons learned, what has gone well, and where there is room for improvement in future development. We hope that this review of SPM-and its aspirations-will provide some context for current imaging neuroscience and foreground some potential directions for the future of the field.

Parkinson's disease is linked to increased beta oscillations in the subthalamic nucleus, which correlate with motor symptoms. However, findings across studies have varied. Our standardized analysis of multicenter datasets reveals that insufficient sample sizes contributed to these discrepancies - a challenge we address by pooling datasets into one large cohort (n=119). Moving beyond beta power, we disentangled spectral components reflecting distinct neural processes. Combining aperiodic offset, low beta, and low gamma oscillations explained significantly more variance in symptom severity than beta alone. Moreover, interhemispheric within-patient analyses showed that, unlike beta oscillations, aperiodic broadband power - likely reflecting spiking activity - was increased in the more affected hemisphere. These findings identify aperiodic broadband power as a potential biomarker for adaptive deep brain stimulation and provide novel insights into the relationship between subthalamic hyperactivity and motor symptoms in human Parkinson's disease. ### Competing Interest Statement Competing interests are currently being requested from all coauthors and will be added once received.

[This corrects the article DOI: 10.3389/fnhum.2025.1544994.].

Statistical Parametric Mapping (SPM) is an integrated set of methods for testing hypotheses about the brain's structure and function, using data from imaging devices. These methods are implemented in an open source software package, SPM, which has been in continuous development for more than 30 years by an international community of developers. This paper reports the release of SPM 25.01, a major new version of the software that incorporates novel analysis methods, optimisations of existing methods, as well as improved practices for open science and software development.

Most scientists need software to perform their research (Barker et al., 2020;Carver et al., 2022;Hettrick, 2014;Hettrick et al., 2014;Switters & Osimo, 2019), and neuroscientists are no exception. Whether we work with reaction times, electrophysiological signals, or magnetic resonance imaging data, we rely on software to acquire, analyze, and statistically evaluate the raw data we obtain-or to generate such data if we work with simulations. In recent years, there has been a shift toward relying on free, open-source scientific software (FOSSS) for neuroscience data analysis (Poldrack et al., 2019), in line with the broader open science movement in academia (McKiernan et al., 2016) and wider industry trends (Eghbal, 2016). Importantly, FOSSS is typically developed by working scientists (not professional software developers), which sets up a precarious situation given the nature of the typical academic workplace wherein academics, especially in their early careers, are on short- and fixed-term contracts. In this paper, we argue that the existing ecosystem of neuroscientific open-source software is brittle, and discuss why and how the neuroscience community needs to come together to ensure a healthy software ecosystem to the benefit of all.

BACKGROUND:Parkinson's disease is linked to increased beta rhythms (13-30 Hz) in the subthalamic nucleus, which correlate with motor symptoms. However, findings across studies are inconsistent. Furthermore, the contribution of other frequencies to symptom severity remains underexplored. METHODS:We analysed subthalamic local field potentials from 119 patients with Parkinson's disease (31 female; mean age 60 ± 9 years) across five independent datasets. Power spectra were parametrised and studied in relation to Levodopa administration and the severity of motor symptoms. FINDINGS:Our findings suggest that small sample sizes contributed to the variable correlations between beta power and motor symptoms reported in previous studies. Here, we demonstrate that more than 100 patients are required for stable replication. Aperiodic offset and low gamma (30-45 Hz) oscillations were negatively correlated with motor deficits (rOffset=-0.32, p=4e-4; rLγ=-0.21, p=0.021), whereas low beta oscillations were positively correlated (rLβ=0.24, p=0.010). Combining offset, low beta, and low gamma power (rLin.reg.(Offset,Lβ,Lγ)=0.47, p=1e-4) explained significantly more variance in symptom severity than low beta alone (J-test: p=2e-5). Interhemispheric within-patient analyses showed that, unlike beta oscillations, aperiodic broadband power (2-60 Hz)-likely reflecting spiking activity-was increased in the more affected hemisphere (Levodopa off-state: p=0.015; on-state: p=0.005). INTERPRETATION:Spectral features beyond conventional beta rhythms are critical to understanding Parkinson's pathophysiology. Aperiodic broadband power shows potential as a new biomarker for adaptive deep brain stimulation, providing important insights into the relationship between subthalamic hyperactivity and motor symptoms in Parkinson's disease. FUNDING:This work was supported by Deutsche Forschungsgemeinschaft (German Research Foundation) Project ID 424778381 TRR 295 "ReTune". H.A. is supported by NIHR UCLH BRC. This work was supported by an MRC Clinician Scientist Fellowship (MR/W024810/1) held by A.O. W.-J.N. received funding from the European Union (ERC, ReinforceBG, project 101077060). E.F. received funding from the Volkswagen foundation (Lichtenberg program 89387). G.W. and L.R. received funding from Deutsche Forschungsgemeinschaft Project ID 511192033.

The cortico-basal ganglia network in Parkinson’s disease (PD) is characterised by the emergence of transient episodes of exaggerated beta frequency oscillatory synchrony known as bursts. Although beta bursts of prolonged duration and amplitude are well recognised to have a detrimental effect on motor function in PD, the neurophysiological mechanisms leading to burst initiation remain poorly understood. Related to this is the question of whether there exist features of basal ganglia activity which can reliably predict the onset of beta bursts. Current state-of-the-art adaptive Deep Brain Stimulation (aDBS) algorithms for PD involve the reactive delivery of stimulation following burst detection and are unable to stimulate proactively so as to prevent burst onset. The discovery of a predictive biomarker would allow for such proactive stimulation, thereby offering further potential for improvements in both the efficacy and side effect profile of aDBS.Here we use deep neural networks to address the hypothesis that beta bursts can be predicted from invasive subthalamic nucleus (STN) recordings in PD patients. We developed a neural network which was able to predict bursts 31.6ms prior to their onset, with a high sensitivity and a low false positive rate (mean performance metrics: sensitivity = 84.8%, precision = 91.5%, area under precision recall curve = 0.87 and false positive rate = 7.6 per minute). Furthermore, by considering data segments that our network labelled as being predictive, we show that a dip in the beta amplitude (a fall followed by a subsequent rise) is a predictive biomarker for subsequent burst occurrence.Our findings demonstrate proof-of-principle for the feasibility of beta burst prediction and inform the development of a new type of intelligent DBS approach with the capability of stimulating proactively to prevent beta burst occurrence.### Competing Interest StatementThe authors have declared no competing interest.

Magnetoencephalography (MEG) recordings are often contaminated by interference that can exceed the amplitude of physiological brain activity by several orders of magnitude. Furthermore, the activity of interference sources may spatially extend (known as source leakage) into the activity of brain signals of interest, resulting in source estimation inaccuracies. This problem is particularly apparent when using MEG to interrogate the effects of brain stimulation on large-scale cortical networks. In this technical report, we develop a novel denoising approach for suppressing the leakage of interference source activity into the activity representing a brain region of interest. This approach leverages spatial and temporal domain projectors for signal arising from prespecified anatomical regions of interest. We apply this denoising approach to reconstruct simulated evoked response topographies to deep brain stimulation (DBS) in a phantom recording. We highlight the advantages of our approach compared to the benchmark-spatiotemporal signal space separation-and show that it can more accurately reveal brain stimulation-evoked response topographies. Finally, we apply our method to MEG recordings from a single patient with Parkinson's disease, to reveal early cortical-evoked responses to DBS of the subthalamic nucleus.