283

Papers

10/10

Roadblocks Active

2

Connections

This study used EEG recordings from 20 healthy adults to model the brain as an energy landscape during emotional face processing, finding that sad states produce significantly more stable (lower-energy) brain network configurations than happy states, especially in the alpha band (effect size d=0.83). The key insight is that stable, hyperconnected networks — a pattern seen in rumination and depression — may have a measurable energy signature that could serve as a state biomarker. If validated in clinical populations, this approach could provide a low-cost, objective index of depressive network states that goes beyond symptom questionnaires.

██████████ 0.9 depression-biomarkers Preprint

ReadSaveConnections

A sparse hierarchical Bayesian model applied to Implicit Association Test reaction times and EEG data achieved AUC of 0.73 for detecting suicidality-related associations and 0.76 for psychosis-related ones — far above the 0.50–0.53 of standard scoring methods; restricting to MDD patients pushed suicidality detection to AUC 0.79. This matters because IATs measure automatic, pre-reflective mental associations that patients may not report voluntarily or consciously, making them a potential complement to self-report screening for suicide risk. The Bayesian approach extracts richer signal from the same behavioral data by treating response parameters as latent variables rather than collapsing them into a single D-score.

██████████ 0.9 computational-psychiatry Preprint

ReadSaveConnections

In a structural MRI study of 222 university students, higher functional AI use (for tasks and learning) was associated with larger dorsolateral prefrontal and hippocampal network efficiency alongside better academic grades, while more frequent socio-emotional AI use (for companionship or emotional support) was associated with higher depression and social anxiety scores. The dissociation suggests the same technology produces opposite brain and mental health profiles depending on the relational role it fills — a tool versus a substitute companion. This is directly relevant to youth mental health policy and to the design of digital therapeutics that incorporate conversational AI agents.

██████████ 0.8 youth-mental-health-crisis Preprint

ReadSaveConnections

The DRIFT method combines factor analysis with adversarial optimization to estimate which treatment will work best for an individual patient even when outcomes are measured across multiple symptom domains and some relevant domains were not measured at all. Applied to the EMBARC sertraline-versus-placebo RCT for major depression, the approach identifies treatment responders more robustly than standard methods by explicitly hedging against unmeasured symptom dimensions like cognitive side effects or somatic burden. For treatment-resistant depression, where patients present with complex multi-domain profiles and have already failed simpler treatment algorithms, this kind of robust individualized prediction could meaningfully improve next-step treatment selection.

██████████ 0.8 treatment-resistant-depression Preprint

ReadSaveConnections

SynSym uses large language models to generate synthetic text describing psychiatric symptoms — expanding each symptom into sub-concepts and generating both clinical and colloquial phrasings — producing training data that performs comparably to real patient text for symptom identification tasks. The practical value is that labeled psychiatric NLP datasets are scarce and ethically difficult to share, and this pipeline could enable teams without clinical data access to build and validate symptom-detection models. Combining synthetic pretraining with small amounts of real data further improved performance, suggesting a practical path for low-resource clinical NLP development.

██████████ 0.7 computational-psychiatry Preprint

ReadSaveConnections

This theoretical paper proposes formalizing neural plasticity as the ratio of the number of nodes in a network to the summed strength of their connections — a simple formula that predicts an optimal intermediate connectivity range where the system is neither too rigid nor too chaotic to change. The idea that psychopathology might correspond to specific deviations from this optimal plasticity range offers a computational framing for why interventions like psychedelics or intensive therapy might restore flexibility in treatment-resistant conditions. The framework is purely theoretical with no new data, so its value lies in generating testable predictions rather than providing immediate clinical tools.

██████████ 0.7 neuroplasticity-interventions Preprint

ReadSaveConnections

This paper proposes reformulating questions about what would happen to brain connectivity if a specific region were disrupted as an energy-perturbation problem, using Hodge theory to separate directed brain communication into components that either dissipate or persist across network cycles. Standard causal methods like Granger causality cannot model cyclic feedback loops common in brain networks, which limits their usefulness for planning interventions like TMS or deep brain stimulation. By framing network intervention as a counterfactual energy calculation, this approach could help predict which brain targets would most efficiently shift a pathological network state — relevant for designing neuromodulation therapies in depression.

██████████ 0.7 neuroplasticity-interventions Preprint

ReadSaveConnections

This review synthesizes progress in transcranial focused ultrasound (tFUS) as a brain stimulation and recording technology, noting it achieves millimeter spatial precision and can target deep structures like the amygdala or subgenual cingulate without the depth-versus-precision tradeoff that limits electromagnetic methods like TMS. For treatment-resistant depression, where deep structures implicated in mood regulation are key targets, tFUS represents a non-invasive alternative to deep brain stimulation that could be integrated into closed-loop systems adjusting stimulation in real time based on neural feedback. The field is review-stage rather than clinical-stage, but the convergence of miniaturization and real-time feedback architectures described here marks meaningful technical maturation.

██████████ 0.7 neuroplasticity-interventions Preprint

ReadSaveConnections

TAAC uses adversarial subspace decomposition to separate depression-relevant acoustic features from speaker identity information in voice recordings, addressing the privacy risk that audio-based depression detection systems could also be used to identify or track individuals without consent. The framework maintains competitive depression detection accuracy while stripping out the identity signal, which is a meaningful step toward making voice biomarker systems viable in real-world clinical and consumer settings where regulatory and ethical scrutiny around biometric data is increasing. This is particularly relevant for scalable passive monitoring applications in digital therapeutics.

██████████ 0.6 depression-biomarkers Preprint

ReadSaveConnections

This theoretical synthesis argues that noise, heterogeneity, and irregular connectivity in biological neural circuits are functional features selected by evolution rather than engineering imperfections — and that computational psychiatry models which smooth these out may be systematically missing the mechanisms that allow brains to adapt. The argument has indirect clinical relevance: if stochastic variability in neural circuits is what enables learning and recovery, then interventions that over-regularize network activity (certain medications, ablative procedures) might inadvertently reduce adaptive capacity. The paper is speculative with no new data, but it provides a useful conceptual challenge to deterministic models of plasticity and recovery.

██████████ 0.6 neuroplasticity-interventions Preprint

ReadSaveConnections