266

Papers

10/10

Roadblocks Active

1

Connections

In 87 MDD patients, the rhythmic switching of the default mode network (a brain circuit tied to self-referential thinking) predicted behavioral symptoms — but only when blood-based inflammatory markers like IL-17A and IL-8 were taken into account. This means inflammation does not simply add to depression; it fundamentally changes the brain-behavior relationship, pointing to a biologically distinct inflammatory subtype of MDD. For clinical trials, this suggests that failing to stratify patients by inflammation status may explain why many antidepressant trials show weak average effects.

██████████ 0.9 depression-biomarkers Peer-reviewed

Read

The standard scoring method for implicit association tests (IATs used to assess suicidality or psychosis risk) performs near chance as a classifier (AUC ~0.50–0.53); a sparse hierarchical Bayesian model integrating EEG, eye tracking, and facial signals from the same task reaches AUC ~0.76 and improves to 0.79 in MDD patients specifically. The model achieves this by treating the IAT as a multivariate signal rather than a single score, and applies structured sparsity suited to small clinical cohorts. This matters because passive biosignal collection during routine psychological assessments could eventually augment or partially replace unreliable self-report for suicide risk screening.

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

ReadSaveConnections

Using a physics-inspired stability model (Hopfield energy) applied to EEG during a face-processing task, sad emotional states produced significantly more stable brain network configurations than happy states — most strongly in the alpha band (Cohen's d = 0.83 in 20 healthy adults). More stable here means the brain settles into a deep attractor state that is harder to exit, which maps directly onto the clinical observation that depressed patients struggle to disengage from sad mood. If replicated in clinical populations, this energy metric could serve as an objective, continuous biomarker for emotional rigidity in depression.

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

ReadSaveConnections

DRIFT addresses a core challenge in precision psychiatry: antidepressant trials measure dozens of symptom items, but standard methods for identifying who will respond require prespecifying which symptoms matter. DRIFT uses generalized factor analysis to extract latent symptom constructs, then estimates individualized treatment effects that are robust to symptoms underrepresented or missing from the trial data — validated on the EMBARC sertraline trial for MDD. The practical payoff is a causal framework for treatment personalization that does not depend on the researcher's choice of which outcomes to emphasize.

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

ReadSaveConnections

Building NLP tools to detect psychiatric symptoms from text requires large annotated patient datasets, which are rare and ethically sensitive. SynSym uses a four-stage large language model pipeline — expanding clinical symptom concepts into colloquial expressions and composing realistic multi-symptom texts based on known co-occurrence patterns — and models trained purely on its output match real-data-trained models on three public benchmarks (D2S, PRIMATE, PsySym). Fine-tuning on even a small amount of real data after synthetic pretraining improves performance further, offering a practical path to building symptom-detection tools for low-resource settings or rare disorders.

██████████ 0.8 depression-biomarkers Preprint

ReadSaveConnections

AI depression-detection models trained on clinical interview transcripts can classify depressed vs. control participants at high accuracy using only the interviewer's scripted questions — not the patient's words — because interviewer prompts are fixed and positionally consistent across sessions. This artifact was found across three datasets (ANDROIDS, DAIC-WOZ, E-DAIC) and both transformer and graph-based architectures, meaning it is systemic, not model-specific. Published benchmark scores in this field likely substantially overstate real-world diagnostic performance, and the field needs evaluation standards that isolate patient-only speech.

██████████ 0.8 depression-biomarkers Preprint

ReadSaveConnections

The relationship between the brain's physical wiring (structural connectivity) and its activity patterns (functional connectivity) is not one-to-one — it operates across multiple spatial scales and hierarchical levels that current models ignore. This framework learns community structure at multiple scales using a differentiable modularity objective, then models cross-hierarchy interactions between structural and functional networks, outperforming state-of-the-art methods on brain age prediction, cognitive score prediction, and disease classification. Better integration of SC-FC relationships could accelerate biomarker discovery for disorders where structure-function mismatch is central, including depression and schizophrenia.

██████████ 0.8 computational-psychiatry Preprint

ReadSaveConnections

"Plasticity" is frequently cited as the mechanism behind antidepressants and psychedelics but is rarely defined precisely enough to be empirically tested. This theoretical paper proposes formalizing plasticity as the ratio of network size to total connection strength: too little plasticity locks a system into rigid patterns (as in depression or PTSD), while too much prevents coherent thought (as in psychosis). The framework is purely conceptual with no new experimental validation, but it provides a testable, quantitative definition that could bridge computational models with clinical intervention design.

██████████ 0.8 neuroplasticity-interventions Preprint

ReadSaveConnections

Standard tools for mapping brain communication (Granger causality, structural equation models) describe correlations but cannot ask "what would happen if this region were perturbed?" — a question central to understanding how therapies like TMS or ketamine work. This paper builds a counterfactual framework that models therapeutic interventions as energy perturbations on brain network flows, using Hodge theory to separate stable versus transient communication patterns, tested on 400 HCP fMRI participants. The approach is methodologically novel but limited by near-zero group-average connectivity in the dataset, so clinical application requires validation in disorder-specific cohorts.

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

ReadSaveConnections

Young adults with schizophrenia scored significantly higher on multiple frailty scales than healthy controls — a finding usually associated with biological aging in elderly populations. Elevated memory γ/δ T cells and Th17-related immune subtypes were detected in schizophrenia patients and correlated with frailty burden, linking immune dysregulation directly to accelerated biological aging. This immune-frailty connection could help explain the well-documented but poorly understood excess physical morbidity and early mortality in psychosis, and suggests immune profiling may have prognostic value beyond symptom scales.

██████████ 0.7 neuroinflammation Peer-reviewed

Read