PLoS Computational Biology
June 12, 2008
David Balduzzi, Giulio Tononi
433 citations
A new time- and state-dependent measure of integrated information, phi, quantifies how much information is generated when a system enters a particular state through causal interactions among its elements, beyond what its parts generate independently. This captures two key properties of consciousness: a large repertoire of experiences that rule out others when one occurs, and the integration of that information into a whole that cannot be decomposed. Applied to discrete networks, phi varies with the state entered, being higher when active and inactive elements are balanced and lower when the network is inactive or hyperactive.
PLoS Computational Biology
August 13, 2009
David Balduzzi, Giulio Tononi
272 citations
Integrated information theory proposes that consciousness corresponds to the amount of integrated information generated by a system of elements, and the quality of an experience is determined by the informational relationships within that system. This paper introduces qualia space (Q), where each possible state of a system is an axis, and submechanisms specify repertoires of states. Arrows between these repertoires define informational relationships that together form a shape—a quale—that uniquely characterizes a conscious experience. The quantity of consciousness is the height of this shape (phi). Entanglement measures how irreducible these relationships are. The framework implies that the same activity pattern can yield different qualia in different systems, and vice versa. Experience cannot be reduced to local mechanisms but requires the entire quale.
PLoS Computational Biology
October 16, 2014
Srivas Chennu, Paola Finoia, Evelyn Kamau et al.
249 citations
Consciousness is thought to require balanced integration and differentiation of brain activity, supported by efficient networks. Analyzing high-density EEG from 32 patients with chronic disorders of consciousness and healthy controls, the authors found that patient networks showed reduced local and global efficiency and fewer hubs in the alpha frequency band. A new metric, modular span, revealed that alpha network modules in patients were spatially limited, lacking the long-distance interactions seen in controls. However, delta and theta band networks were partially reversed and more similar to each other in patients. Alpha network efficiency correlated with behavioral awareness. Notably, some behaviorally unresponsive vegetative patients with covert awareness had well-preserved alpha networks resembling controls, suggesting network mechanisms that may support consciousness despite severe behavioral impairment.
PLoS Computational Biology
January 20, 2011
Adam B. Barrett, Anil K. Seth
243 citations
Two new measures of integrated information, Φ(E) and Φ(AR), overcome limitations of the earlier Φ(DM) measure, which could rarely be applied to biological systems because it required discrete Markov dynamics. The new measures are easy to apply to time-series data, as demonstrated through simulations. They offer new opportunities for studying information integration in real and model systems and have implications for understanding consciousness and other neurocognitive processes. However, the findings also challenge theories that assign physical meaning to these measured quantities.
PLoS Computational Biology
November 7, 2007
Marco Loh, Edmund T Rolls, Gustavo Deco
172 citations
Reduced depth in the basins of attraction of cortical attractor states destabilizes neural activity at the network level due to constant statistical fluctuations from stochastic spiking of neurons. In integrate-and-fire network simulations, decreasing NMDA receptor conductances reduces attractor basin depth, destabilizes short-term memory states, and increases distractibility. Cognitive symptoms of schizophrenia—distractibility, working memory deficits, poor attention—could stem from this instability in prefrontal cortical networks. Lower firing rates in orbitofrontal and anterior cingulate cortex may account for negative symptoms like reduced emotions. Decreasing both GABA and NMDA conductances causes switches between attractor states and jumps from spontaneous activity into attractors, linked to positive symptoms such as delusions, paranoia, and hallucinations from shallow basins in temporal lobe semantic memory networks.
PLoS Computational Biology
April 14, 2015
Joon-Young Moon, UnCheol Lee, Stefanie Blain-Moraes et al.
143 citations
Efficient brain networks balance global integration with functional specialization, but how global topology, local node dynamics, and information flow relate has been unclear. Using analytical solutions of oscillator models, computational simulations on model and anatomical brain networks, and high-density electroencephalography from conscious and anesthetized humans, the authors demonstrate that network nodes with more connections (higher degree) have larger amplitudes and are directional targets (phase lag) rather than sources (phase lead). This degree–directionality relationship appears to be a fundamental network property with direct applicability to brain function. Changes in directionality patterns across states of human consciousness are driven by alterations in brain network topology.
PLoS Computational Biology
August 30, 2018
Hyoungkyu Kim, Joon-Young Moon, George A. Mashour et al.
79 citations
Hysteresis—the difference between the forward and reverse paths of state transitions—occurs as people lose and regain consciousness. Analyzing high-density EEG from healthy volunteers given sevoflurane or ketamine, the authors found that functional brain networks exhibit hysteresis during these transitions. The principle of explosive synchronization, which governs abrupt state shifts in many complex networks, also explains hysteresis in the brain. More potent anesthetics produce larger hysteresis; a broader range of EEG frequencies hastens the loss of consciousness but delays its return; connectivity shows greater hysteresis than EEG power; and network structure and strength reconfigure differently during loss versus recovery. These results indicate that hysteresis in conscious state transitions is a generic network feature, potentially allowing prediction and modulation of such transitions.
PLoS Computational Biology
August 2, 2023
Alexander Tscshantz, Beren Millidge, Anil K. Seth et al.
56 citations
Predictive coding theory holds that the brain perceives by minimizing prediction errors through cycles of neural activity. However, some visual perception, including complex object recognition, happens too quickly for such cycles. This paper proposes that the initial fast 'feedforward sweep' performs amortized inference, using a learned function to map data directly to beliefs, while slower recurrent processing performs iterative inference, sequentially updating beliefs for greater accuracy. A hybrid predictive coding network combining both methods is introduced, implemented in a biologically plausible neural architecture using local Hebbian rules. The hybrid model achieves rapid perception for familiar data while retaining context-sensitivity and sample efficiency for novel situations, and adaptively balances both inference modes based on uncertainty.
PLoS Computational Biology
August 18, 2021
Indie C. Garwood, S. Chakravarty, Jacob Donoghue et al.
32 citations
Ketamine, an anesthetic that blocks NMDA receptors, produces alternating bursts of gamma (25-50 Hz) and slow-delta (0.1-4 Hz) brain oscillations. A hidden Markov model fitted to local field potentials from two non-human primates and electroencephalograms from nine humans quantified these dynamics. Gamma activity lasted on average 2.2 seconds in one primate, 1.2 in the other, and 2.5 in humans; slow-delta lasted 1.6, 1.0, and 1.8 seconds respectively. Five sub-states with regular sequential transitions were identified. These findings provide quantitative constraints for models of rhythm generation underlying ketamine-induced altered arousal.
PLoS Computational Biology
February 1, 2023
Giulio Ruffini, Giada Damiani, Diego Lozano-Soldevilla et al.
28 citations
Brain dynamics under LSD become more disordered and complex, moving further from the critical point that characterizes healthy brain function. Using Ising spin models fitted to fMRI data from fifteen participants, the authors show that LSD reduces interhemispheric connectivity, especially between corresponding regions in opposite hemispheres. Ising temperatures were significantly higher under LSD than placebo, indicating a shift into a more disordered (paramagnetic) state. Algorithmic complexity of brain activity, measured by block decomposition, correlated with both Ising temperature and condition, supporting the entropic brain hypothesis that psychedelics increase neural disorder.
PLoS Computational Biology
August 24, 2017
Peter J. Hellyer, Claudia Clopath, Angie A. Kehagia et al.
21 citations
A simple computational model of spontaneous neural dynamics controlling an agent in a virtual environment shows that brain-environment feedback can rapidly destabilize neural and behavioral dynamics, requiring homeostatic mechanisms. Local homeostatic plasticity, where inhibition adjusts to balance excitation, and global mechanisms, where regional task-negative activity compensates for task-positive sensory input in another region, both stabilize behavior. The results suggest complementary functional roles for local and macroscale homeostatic processes and propose a novel function for macroscopic task-negative activity patterns, such as the default mode network, in maintaining stable neural and behavioral dynamics.
PLoS Computational Biology
June 6, 2025
Wiep Stikvoort, Eider Pérez-Ordoyo, Iván Mindlin et al.
3 citations
A person's level of consciousness can be assessed by how the brain reacts to stimulation, but this study shows that the brain's unperturbed activity already contains that information. Using personalized whole-brain models fitted to resting-state fMRI data from people in altered states of consciousness (deep sleep, disorders of consciousness), the researchers measured the brain's out-of-equilibrium dynamics—specifically, the asymmetry of effective connections and time irreversibility. They found that states with lower arousal or awareness had less asymmetric connectivity, less irreversibility, and lower complexity in simulated responses compared to controls. The asymmetry in connections drives the nonequilibrium state and, in turn, the differences in complexity.
PLoS Computational Biology
June 9, 2025
Jessie Rademacher, Tineke Grent-'t-Jong, Davide Rivolta et al.
2 citations
Ketamine, an NMDA receptor antagonist given at sub-anesthetic doses, flattens the aperiodic slope of brain activity and increases gamma-band power (30–90 Hz), especially in prefrontal and central regions. These effects correlate with gene expression of parvalbumin and GluN2D. A computational model of cortical layer 2/3 shows that reducing NMDA receptor activity in parvalbumin or somatostatin interneurons boosts pyramidal neuron firing, reproducing the gamma power increase but not the aperiodic slope change. This suggests parvalbumin and somatostatin interneurons drive the gamma power rise, while the aperiodic component involves other mechanisms, challenging current excitation/inhibition balance models.
PLoS Computational Biology
October 28, 2025
Fernando Lehue, Carlos Coronel-Oliveros, Vicente Medel et al.
1 citation
During sleep, brain dynamics shift from wakefulness through NREM stages N1, N2, and N3, driven partly by decreases in the neuromodulators acetylcholine (ACh) and noradrenaline (NA). Analyzing fMRI data from healthy individuals and using a whole-brain model, the study shows that functional connectivity (FC) changes distinctly: locus coeruleus connectivity with the cortex decreases during N2 and N3, while basal forebrain connectivity with the cortex decreases during N3. Compared to wakefulness, the brain becomes more integrated in N1 and more segregated in N3. Region-specific neurotransmitter effects are key to explaining these FC changes, advancing understanding of how neurochemistry modulates sleep stages and consciousness transitions.
PLoS Computational Biology
December 15, 2025
Lingyu Zhang, Weiyang Shi, Ziyang Zhao et al.
LSD increases global brain synchrony and dynamic complexity by stabilizing a globally synchronized, functionally non-modular brain state that acts as an attractor, recruiting transitions from cognitive control networks. This enhanced synchrony arises from a convergence of excitatory/inhibitory balance across cortical hierarchies, driven by suppression in sensorimotor cortices and potentiation in transmodal regions. Sensorimotor cortices emerge as potential regulatory hubs for this rebalancing. The resulting brain state shows weakened sensory anchoring and enhanced cognitive flexibility, blurring the line between concrete perception and abstract cognition. This neurophysiological remodeling may underlie LSD's hallucinatory effects and its therapeutic potential for mental disorders with rigid thought patterns.
PLoS Computational Biology
March 1, 2026
Miłosz Danilczuk, Marek Pokropski, Piotr Suffczynski
Integrated Information Theory (IIT) proposes that consciousness arises from a system's ability to integrate information. Applying IIT to a simulated network of integrate-and-fire (IAF) neurons shows that such a network can have a non-zero Φ value—a measure of integrated information—under specific conditions. The complexity of the network's dynamics does not necessarily correlate with its Φ value. However, the amount of integrated information increases with the neurons' time constant, reflecting their integrative capacity. The integrated information measure defined in IIT 3.0 is not resilient to noise when the network includes internal random fluctuations.
PLoS Computational Biology
February 1, 2026
Tao Xia, Chuan-Peng Hu, Başak Türker et al.
Sleep is often seen as a state of mental disconnection, but new research shows that decision-making abilities can persist during certain sleep stages. Using facial muscle measurements and computational modeling, the study found that people could make lexical decisions (distinguishing real words from pseudowords) during N1 sleep and lucid REM sleep, though through different brain processes. During N1 sleep, both enhanced sensory-motor processing and increased evidence accumulation supported decisions about words, while pseudowords were selectively impaired, suggesting cognitive resources prioritize meaningful stimuli. In lucid REM sleep, decisions relied solely on evidence accumulation, and participants raised their decision thresholds, requiring more evidence before responding to maintain accuracy despite reduced efficiency. Sleep involves dynamic, state-specific reconfiguration of decision-making mechanisms rather than passive cognitive decline.
PLoS Computational Biology
September 1, 2025
Jordy Peeters, Dimitri de Bundel, Kenno Vanommeslaeghe
The serotonin-2A receptor (5-HT2AR) is a target for antidepressants that could work quickly or in treatment-resistant cases, but activating it can cause hallucinations. Recent research suggests certain partial agonists might produce antidepressant effects without hallucinations, though the molecular details are unclear. This study used molecular dynamics simulations of the receptor bound to two antipsychotics, three potential non-hallucinogens, and two hallucinogens. Findings suggest modest receptor activation yields only antidepressant effects, while hallucinations result from excessive activation. Modest activation via a sufficiently weak partial agonist may offer a viable drug development pathway, whereas microdosing may be problematic due to abuse potential and narrow therapeutic windows.
PLoS Computational Biology
August 1, 2024
Alireza Zaeemzadeh, Giulio Tononi
Integrated information theory, originally a theory of consciousness, offers a mathematical way to measure how causally irreducible a system or subset of its units is. Mechanism integrated information quantifies how much of a mechanism's causal power cannot be explained by its parts; if fully explained by its parts, integrated information is zero. This work studies the upper bound of this measure and how it is achieved, examining mechanisms in isolation, groups of mechanisms, and groups of causal relations among them. New theoretical results show that mechanisms sharing parts cannot all reach their maximum simultaneously. Techniques are introduced to design systems that maximize integrated information for subsets of mechanisms or relations, potentially reducing computations and comparing connectivity profiles.
PLoS Computational Biology
October 1, 2023
Christoffer Lundbak Olesen, Peter Thestrup Waade, Larissa Albantakis et al.
Two major theoretical frameworks—the Free Energy Principle, which describes how self-organizing systems maintain order, and Integrated Information Theory, which aims to mathematically characterize conscious experience—are brought together. Analyzing data from an earlier evolutionary simulation, the authors show that agents' surprisal (a measure from the Free Energy Principle) decreases as their fitness and neural integration increase over evolutionary time. Furthermore, surprisal fluctuates in tandem with IIT-based measures of consciousness within individual trials. This suggests that IIT's consciousness measures are partly shaped by the agent's interaction with its environment, making a formal link between the two theories possible.
PLoS Computational Biology
July 1, 2022
Michelle J. Redinbaugh, Mohsen Afrasiabi, Jessica M. Phillips et al.
Deep brain stimulation (DBS) of the central lateral thalamus in macaques can produce episodes of vacant staring with low-frequency brain oscillations, termed vacant, perturbed consciousness (VPC). The likelihood of VPC depended on stimulation frequency. During VPC, measures of neural complexity and integration decreased, and communication in cortico-striato-thalamic circuits changed substantially, with increased low-frequency power and coherence, especially in thalamo-parietal and cortico-striatal pathways. These features resembled absence epilepsy. The same DBS method, at different frequencies, can also increase consciousness in anesthetized macaques, offering a flexible tool to study consciousness with fewer confounds and to inform clinical research on consciousness disorders.
PLoS Computational Biology
February 1, 2021
Angus Leung, Dror Cohen, Bruno van Swinderen et al.
Consciousness may arise from integrated patterns of causal interactions among neurons, measurable as an informational structure. In fruit flies, integrated interactions among neuronal populations during wakefulness collapsed into isolated clusters under anesthesia. Informational structures distinguished wakeful from anesthetized states more accurately than a simpler scalar measure. Rich information structures, which cannot arise from purely feedforward systems, occurred across the fly brain and collapsed uniformly during anesthesia. The concept of an informational structure may serve as a useful measure for level of consciousness.
PLoS Computational Biology
September 1, 2018
Francisco J. Esteban, Javier A. Galadí, José A. Langa et al.
Integrated Information Theory (IIT) offers a mathematical framework for consciousness, identifying conscious experience with a conceptual structure that is composed of parts, informative, integrated, and maximally irreducible. This paper extends IIT by introducing a space-time continuous version of integrated information. Using graph and dynamical systems approaches, it defines an Informational Structure for a mechanism in a given state, associated with the system's global attractor over time. This structure determines all past and future behavior, enriches phase space points with cause-effect power via an Informational Field, and allows a measure of integrated information through invariants and transition probability matrices.
PLoS Computational Biology
June 10, 2016
Masafumi Oizumi, Shun-ichi Amari, Toru Yanagawa et al.
A novel practical measure called Φ* quantifies integrated information in the brain, a property predicted by Integrated Information Theory (IIT) to reflect levels of consciousness. Earlier measures failed to satisfy theoretical lower and upper bounds: zero when no information is generated or when parts are independent, and the total information generated by the whole system. By applying mismatched decoding from information theory, Φ* meets these bounds. Under a Gaussian assumption, Φ* has an analytical expression applicable to experimental neural data. Φ* can serve as a measure of integrated information in consciousness research and as a tool for network analysis in biology.
PLoS Computational Biology
October 1, 2008
Anandamohan Ghosh, Y Rho, A R McIntosh et al.
Resting state networks, which are collections of brain regions showing coordinated activity even when no task is being performed, can emerge from a stability analysis of network dynamics using biologically realistic primate brain connectivity. Anatomical information alone does not identify these networks; noise and time delays from signal propagation along connecting fibers are essential for their emergence. The spatiotemporal dynamics operate on multiple time scales, producing both fast neuroelectric oscillations (1–100 Hz) and slow hemodynamic oscillations (<0.1 Hz). The combination of structure and time delays creates a space-time framework where neural noise allows the brain to explore various functional configurations.