Edmundo Lopez-Sola

Objective: Transcranial direct current stimulation (tDCS) is a promising non-invasive therapy for drug-resistant epilepsy, but the montage optimization pipelines used to personalize it often rely on biophysical head models that capture only the induced electric field and ignore the network dynamics through which stimulation influences epileptogenic activity. We aimed to develop and evaluate a montage optimization framework that incorporates individualized network dynamics. Approach: We introduce a framework based on a digital twin, or neurotwin , a personalized replica of a patient’s brain that couples a biophysical head model of the stimulation-induced field with a personalized whole-brain model (WBM) of seizure dynamics constrained by structural connectivity and intracranial recordings. We applied it to produce optimal montages in 12 patients with drug-resistant epilepsy, compared them with conventional biophysics-based optimization, and assessed the framework retrospectively against the clinical outcomes of six treated patients. Main results: Because the neurotwin optimization suppresses simulated spread by construction, the informative result is the divergence between pipelines: the network-informed and field-based optimizations produced substantially different montages despite delivering comparable inhibitory fields at the epileptogenic zone, and montages with nearly identical field distributions could yield markedly different network outcomes within the model. A minimum-replacement-set analysis traced these differences to field changes at a few propagation-zone and off-target parcels rather than to the overall field. The parcels whose inhibition most reduced spread tended to occupy highly connected (hub) positions in the structural connectome, a robust but partial association indicating that network structure shapes, but does not fully determine, where stimulation is most effective. In the retrospective analysis, model-predicted spread reduction tracked clinical seizure-frequency change in the expected direction, though not significantly ( n = 6). Significance: We hypothesize that accounting for individualized network dynamics may improve seizure control, a hypothesis now being tested prospectively ( NCT06334952 ). Highlights: A neurotwin couples a biophysical head model with a personalized whole-brain model (WBM) for tDCS optimization. neurotwin and field-based optimization yield divergent montages despite comparable EZ fields. Field differences at a few target and off-target parcels, not overall field magnitude, drive the simulated network outcomes. Parcels controlling simulated seizure spread tend to be structural network hubs (high nodal strength and centrality), an association that persists after adjusting for local dynamics. WBM-predicted spread reduction was directionally consistent with clinical seizure-frequency change (non-significant, n = 6).

Classical serotonergic psychedelics show promise in addressing neurodegenerative disorders such as Alzheimer’s disease by modulating pathological brain dynamics. However, the precise neurobiological mechanisms underlying their effects remain elusive. This study introduces a personalized whole-brain model built upon a laminar neural mass framework to elucidate these effects. Using multimodal neuroimaging data from 30 subjects diagnosed with mild to moderate Alzheimer’s disease, we simulate the impact of serotonin 2A receptor activation, characteristic of psychedelics, on cortical dynamics. By modulating the excitability of layer 5 pyramidal neurons, our models reproduce hallmark changes in EEG power spectra observed under psychedelics, including alpha power suppression and gamma power enhancement. These spectral shifts are shown to correlate strongly with the regional distribution of serotonin 2A receptors. Furthermore, simulated EEG reveals increased complexity and entropy, suggesting restored network function. These findings underscore the potential of serotonergic psychedelics to reestablish healthy oscillatory dynamics in the prodromal and early phases of Alzheimer’s disease and offer mechanistic insights into their potential therapeutic effects in neurodegenerative disorders.

The phenomenon of “pure awareness”, central to many contemplative traditions, has recently attracted scientific interest for its relevance to the study of consciousness. In this paper, we investigate pure awareness through the algorithmic agent model, a computational framework with roots in algorithmic information theory. This framework proposes that agents build compressive models of the world, and that structured experience arises from running such models. By linking phenomenology, dynamical systems theory, and computation, the algorithmic agent model offers a useful framework for investigating experiences with minimal structure or phenomenal content. We propose that pure awareness, as a stand-alone phenomenon, may correspond to experiences with minimal structure achieved through meditation, psychedelics, or other deconstructive practices. To illustrate this, we explore how such experiences may arise through jhāna meditation. A key hypothesis is that the phenomenology of pure awareness arises from a specific model: the agent's model of its own modeling process. We argue that the agent's recognition of the modeling process can occur alongside other phenomenal content, as in non-dual awareness. We also outline how these ideas can be explored through whole-brain computational models based on predictive processing theory, and empirically grounded in meditation and psychedelic research.

The phenomenon of “pure awareness”, central to many contemplative traditions, has recently attracted scientific interest for its relevance to the study of consciousness. In this paper, we investigate pure awareness through the algorithmic agent model, a computational framework with roots in algorithmic information theory. This framework proposes that agents build compressive models of the world, and that structured experience arises from running such models. By linking phenomenology, dynamical systems theory, and computation, the algorithmic agent model offers a useful framework for investigating experiences with minimal structure or phenomenal content. We propose that pure awareness, as a stand-alone phenomenon, may correspond to experiences with minimal structure achieved through meditation, psychedelics, or other deconstructive practices. To illustrate this, we explore how such experiences may arise through jhāna meditation. A key hypothesis is that the phenomenology of pure awareness arises from a specific model: the agent's model of its own modeling process. We argue that the agent's recognition of the modeling process can occur alongside other phenomenal content, as in non-dual awareness. We also outline how these ideas can be explored through whole-brain computational models based on predictive processing theory, and empirically grounded in meditation and psychedelic research.

Early‐stage AD involves cortical hyperexcitability, progressing to oscillatory slowing and hypoactivity. These changes are linked to parvalbumin‐positive () interneuron dysfunction and neuronal loss driven by amyloid‐beta () and hyperphosphorylated tau (hp‐), though underlying mechanisms remain unclear. To investigate this relationship, we employed a Laminar Neural Mass Model integrating excitatory and inhibitory populations. Synaptic coupling from interneurons to pyramidal cells was progressively reduced to mimic ‐induced neurotoxicity. Additional parameter variations simulated alternate mechanisms, including hp‐tau pathology. Simulated dipole activity was analyzed in the time‐frequency domain and compared to the literature. Simulating interneuron dysfunction reproduced AD's biphasic progression: early hyperexcitability with elevated gamma and alpha power, followed by oscillatory slowing and reduced spectral power. Alternative mechanisms, such as increased excitatory drive, did not replicate this trajectory. To account for late‐stage hypoactivity and reduced firing rates, we incorporated pyramidal cell disruption consistent with hp‐ neurotoxicity. While not essential for local oscillatory changes, this addition aligns the model with empirical markers of advanced AD and supports whole‐brain modeling. These findings highlight interneuron dysfunction as a primary mechanism of early electrophysiological disruption in AD, with pyramidal cell loss contributing to late‐stage hypoactivity, offering a mechanistic model for excitation–inhibition imbalance across progression.

Objective. Computational modeling has recently emerged as a powerful tool to better understand seizure dynamics and guide new treatment strategies. This work aims to develop and personalize whole-brain computational models in epilepsy using multimodal clinical data to simulate and evaluate individualized therapeutic strategies. Approach. We present a computational framework that constructs patient-specific whole-brain models of seizure propagation by integrating SEEG, MRI, and diffusion MRI data. The pipeline uses neural mass models for each node in the network, simulating whole-brain dynamics. Model personalization involves adjusting global and local parameters representing the excitability of individual brain areas, using an evolutionary algorithm that aims to maximize the correlation between empirical and synthetic functional connectivity matrices derived from SEEG data. Main results. The resulting personalized models successfully reproduce individual seizure propagation patterns and can be used to simulate therapeutic interventions like surgery, stimulation, or pharmacological interventions within a unified physiological framework. Notably, model predictions reveal distinct patient-specific responses across interventions, including variable sensitivity to different pharmacological agents and identification of critical regions whose removal or modulation reduced seizure spread. Significance. This framework provides a mechanistic, interpretable approach to simulate and compare individualized treatment strategies. By integrating multimodal data into a unified whole-brain model, it has the potential to improve clinical decision-making in epilepsy by identifying accessible and functionally relevant targets.

Minimal phenomenal experiences (MPE) provide a distinctive entry point for investigating consciousness and understanding how certain meditative practices may represent and contribute to human flourishing. Adopting a neurophenomenological stance, we combine first-person phenomenology with formal whole-brain modeling to ask what mechanisms underlie MPE states and the progression towards it, as reached through jhāna meditation, a type of advanced concentrative absorption meditation (ACAM-J). We used a dataset of advanced practitioners performing the eight ACAM-J states scanned in high-field functional magnetic resonance imaging. For each state, we build a mechanistic whole-brain model that reproduces brain activity by modeling the dynamical regimes of different resting-state networks. We found that the later ACAM-J states, taken here as candidates for MPE, show a shift of functional network dynamics toward near-bifurcation working points. The default mode network exhibits the largest shift, from a distant noise-driven regime at a control condition to near-critical dynamics. We also observed that the pathway to MPE is nonlinear, with prominent reconfigurations at key meditative milestones. Our results suggest that MPE corresponds to a globally susceptible state where near-critical dynamics dominate. We interpret this critical regime as a form of “openness”, where the usual hierarchical and heterogeneous patterns of brain activity give way to generalized flexibility. In this light, MPE provides both a minimal model of consciousness and a window into how advanced meditation can profoundly reshape brain dynamics.

Objective: Computational modeling has recently emerged as a powerful tool to better understand seizure dynamics and guide new treatment strategies. This work aims to develop and personalize whole-brain computational models in epilepsy using multimodal clinical data to simulate and evaluate individualized therapeutic strategies. Approach: We present a computational framework that constructs patient-specific whole-brain models of seizure propagation by integrating SEEG, MRI, and diffusion MRI data. The pipeline uses neural mass models for each node in the network, simulating whole-brain dynamics. Model personalization involves adjusting global and local parameters representing the excitability of individual brain areas, using an evolutionary algorithm that aims to maximize the correlation between empirical and synthetic functional connectivity matrices derived from SEEG data. Main results: The resulting personalized models successfully reproduce individual seizure propagation patterns and can be used to simulate therapeutic interventions like surgery, stimulation, or pharmacological interventions within a unified physiological framework. Notably, model predictions reveal distinct patient-specific responses across interventions, including variable sensitivity to different pharmacological agents and identification of critical regions whose removal or modulation reduced seizure spread. Significance: This framework provides a mechanistic, interpretable approach to simulate and compare individualized treatment strategies. By integrating multimodal data into a unified whole-brain model, it has the potential to improve clinical decision-making in epilepsy by identifying accessible and functionally relevant targets.

Alzheimer’s disease (AD) is characterized by a progressive cognitive decline underpinned by disruptions in neural circuit dynamics. Early-stage AD is associated with cortical hyperexcitability, whereas later stages exhibit oscillatory slowing and hypoactivity, a progression observable in electrophysiological spectral characteristics. While previous studies have linked these changes to the dysfunction of fast-spiking parvalbumin-positive ( PV ) interneurons and neuronal loss associated with amyloid-beta (Aβ) and hyperphos-phorylated tau (hp- τ ) pathology, the precise mechanistic relationship between cellular and altered electrophysiology remains unclear. To study this relationship, we employed a Laminar Neural Mass Model that integrates excitatory and inhibitory neural populations within a biophysically informed columnar framework. The connectivity constant from PV cells to pyramidal neurons was gradually reduced to simulate the progressive neurotoxic effects of Aβ oligomers. Other model parameters were systematically varied to compare with existing modeling literature and also to simulate the effects of hp- τ . All model predictions were compared to empirical M/EEG findings in the literature. Our simulations of PV interneuron dysfunction successfully reproduced the biphasic electrophysiological progression observed in AD: an early phase of hyperexcitability with increased gamma and alpha power, followed by oscillatory slowing and reduced spectral power. Alternative mechanisms and model parameters, such as increased excitatory drive, failed to replicate the observed biomarker trajectory. Additionally, to reconcile the hypoactivity and decreased firing rates observed in advanced AD stages, we combined the PV dysfunction model with a disruption of the pyramidal cell populations that reflects the neurotoxicity induced by hp- τ . Although this additional mechanism is not necessary to reproduce oscillatory changes in the isolated neural mass, it is crucial for aligning the model with evidence of reduced firing rates, metabolic activity, and cell loss and will enhance its applicability in future whole-brain modeling studies. These results support the hypothesis that at the local level, PV interneuron dysfunction is a primary driver of cortical electrophysiological alterations, while pyramidal neuron loss underlies later-stage severe hypoactivity. Our model provides a mechanistic framework for interpreting excitation-inhibition imbalance across AD progression, demonstrating the value of biophysically constrained models for interpreting electrophysiological biomarkers. Author summary: Alzheimer’s disease (AD) is not just a disorder of memory—it is a disease of brain networks. Before widespread neuronal loss occurs, the brain enters a state of hyperexcitability, which may contribute to disease progression. This hyperactivity is thought to arise from the selective dysfunction of inhibitory interneurons, particularly parvalbumin-positive ( PV ) interneurons, which play a crucial role in maintaining balanced brain activity. However, as the disease advances, a dramatic shift occurs, with neurons becoming progressively less active, leading to network breakdown. Understanding how this transition unfolds is essential for identifying new targets for early intervention. In this study, we developed a computational model of AD that represents the effects of amyloid-beta (Aβ) oligomers and hyperphosphorylated tau (hp- τ ) on neural circuits in a biologically meaningful way. Our model reproduces key features of M/EEG biomarkers, demonstrating that PV interneuron dysfunction leads to early hyperexcitability and the electrophysiological oscillatory changes characteristic of AD, while pyramidal cell pathology is necessary to drive later hypoactivity and network failure. By bridging molecular pathology mechanisms with mesoscale neural activity, our model provides a powerful tool for studying AD-related circuit dysfunction. It also highlights the importance of PV interneurons as a potential therapeutic target, paving the way for biologically informed interventions that could alter the course of the disease.

Predictive coding frameworks suggest that neural computations rely on hierarchical error minimization, where sensory signals are evaluated against internal model predictions. However, the neural implementation of this inference process remains unclear. We propose that cross-frequency coupling (CFC) furnishes a fundamental mechanism for this form of inference. We first demonstrate that our previously described Laminar Neural Mass Model (LaNMM) supports two key forms of CFC: (i) Signal-Envelope Coupling (SEC), where lowfrequency rhythms modulate the amplitude envelope of higher-frequency oscillations and (ii) Envelope-Envelope Coupling (EEC), where the envelopes of slower oscillations modulate the envelopes of higher-frequency rhythms. Then, we propose that, by encoding information in signals and their envelopes, these processes instantiate a hierarchical “Comparator” mechanism at the columnar level. Specifically, SEC generates fast prediction-error signals by subtracting top-down predictions from bottom-up oscillatory envelopes, while EEC operates at slower timescales to instantiate gating—a critical computational mechanism for precision-weighting and selective information routing. To establish the face validity and clinical implications of this proposal, we model perturbations of these CFC mechanisms to investigate their roles in pathophysiological and altered neuronal function. We illustrate how, in disorders such as Alzheimer’s disease, disruptions in gamma oscillations following dysfunction in fast-spiking inhibitory interneurons impact Comparator function with an aberrant amplification of prediction errors in the early stages and a drastic attenuation in late phases of the disease. In contrast, by increasing excitatory gain, serotonergic psychedelics diminish the modulatory effect of predictions, resulting in a failure to attenuate prediction error signals (c.f., a failure of sensory attenuation). Collectively, these findings implicate cross-frequency coupling across multiple temporal scales as a key computational mechanism supporting predictive coding and suggest that disruptions in these processes play a central role in disease. Highlights: Using an encoding scheme where information is encoded in signals, their envelopes, and envelopes of envelopes, we show how to implement prediction error and precision modulation in a neural mass model through cross-frequency coupling (CFC). We use the laminar neural mass model (LaNMM), which integrates Jansen-Rit and pyramidal interneuron gamma (PING) submodels to display fast and slow rhythms and provides mechanisms for a) Signal-Envelope Coupling (SEC) , where slow-wave activity modulates the amplitude envelope of fast oscillations (analogous to phase-amplitude coupling), and b) Envelope-Envelope Coupling (EEC) , where the envelopes of slower oscillations modulate the envelopes of higher-frequency rhythms. We show how to use the LaNMM to implement information-based prediction-error evaluation (as used in Active Inference and Kolmogorov Theory), computing the approximate precision-weighted difference between incoming sensory data (envelopes) and internal model predictions (signals or envelopes). We show that using these mechanisms, the Comparator mechanism can operate at multiple levels and timescales, generating fast prediction-error signals (via SEC) and slower gating signals that encode context (e.g., precision) (via EEC). Our model provides insights into the physiological and cognitive consequences of mesoscale circuital alterations in the context of predictive coding. First, we study disorders of fast-spiking interneurons, such as Alzheimer’s Disease (AD). In the early stages of AD, error evaluation and precision are disrupted (inflated error and reduced gating/weight of predictions), leading to higher prediction errors. In later stages, prediction errors are suppressed regardless of predictions or their precision. Then, we show how serotonergic psychedelics increase the effective weight of inputs and diminish that of predictions, resulting in higher prediction error signals. These observations link oscillatory mechanisms and predictive coding alterations, and potentially with the subjective phenomena in each condition—including cognitive decline in AD and hallucinatory states under psychedelics.