Biology subjects

Dragomir, A.

Publications and source records attributed to Dragomir, A..

3 recordsLinked to original sources

bioRxiv2026

Probing EEG Foundation Model Attention Maps for Lightweight Interpretable Emotion Recognition

Foundation Models (FMs) trained on EEG signals promise subject- and dataset-independent pipelines for brain-computer interface (BCI) applications, but have yet to consistently outperform task-specific architectures, and their large representations are costly to compute and store. As a means of advancing the understanding of the Foundation Models' inner mechanisms, we propose a lightweight alternative that exploits the self-attention mechanism of REVE, a pretrained EEG FM, to derive features similar to weighted Functional Connectivity (FC) without any fine-tuning. Inspired by kernel-based FC matrices, we aggregate the raw first-layer attention weights of a single head over the full duration of dataset trials into weighted channel-to-channel adjacency matrices. These matrices replace standard encoder representation vectors as classification features, substantially reducing computational cost. We further introduce a label-free, symmetry-based criterion, motivated by the emergence of near-symmetric attention patterns in trained encoder-decoder transformers, ranking attention heads by how much discriminative information they carry. Classical machine learning classifiers are then trained on the single most informative head's attention-derived FC matrix and achieve balanced accuracy and F1 scores on the SEED-IV emotion-recognition dataset that are competitive with several fully fine-tuned state-of-the-art EEG Foundation Models, requiring at least two orders of magnitude fewer calculations than the standard use of the encoder. Subsequent cortical region analysis revealed a significant over-representation of frontal region connections by ranking the most discriminative FC matrix connectivity edges, validating our methods as consistent with prior affective-computing literature. These results suggest that interpretable, computationally inexpensive linear probing of attention maps is a viable route to deploying EEG Foundation Models in resource-constrained settings.

neuroscience↗

bioRxiv2026

Excitatory delay-coupling explains in-phase and antiphase functional connectivity

Coordinated oscillatory activity between brain regions underpins cognition, yet the phase relationships governing this coordination remain poorly understood. Using scalp EEG from 31 participants performing a motor learning task, we show that inter-site phase clustering occurs predominantly at in-phase or antiphase relationships, with the transition between them governed by conduction delay. Homologous interhemispheric pairs remain in-phase despite long distances, consistent with faster callosal conduction. A minimal model of two delay-coupled excitatory populations reproduces these features without parameter tuning; stability analysis shows that in-phase and antiphase oscillations arise from competing instabilities, with delay determining which dominates. Task performance shifts a frontoparietal network toward in-phase connectivity, a modulation captured by the Phase Relationship Index (PRI) but missed by conventional clustering metrics. Preliminary evidence from independent datasets suggests these patterns generalise. These findings challenge the assumption that zero-lag connectivity reflects volume conduction, offer a mechanistic account linking conduction delay to phase organisation, and introduce PRI as a metric sensitive to functionally relevant connectivity changes invisible to existing measures.

neuroscience↗

bioRxiv2020

Rare variants in dynein heavy chain genes in two individuals with situs inversus and developmental dyslexia

BackgroundDevelopmental dyslexia (DD) is a neurodevelopmental learning disorder with high heritability. A number of candidate susceptibility genes have been identified, some of which are linked to the function of the cilium, an organelle regulating left-right asymmetry development in the embryo. Furthermore, it has been suggested that disrupted left-right asymmetry of the brain may play a role in neurodevelopmental disorders such as DD. MethodsHere, we studied two individuals with co-occurring situs inversus (SI) and DD using whole genome sequencing to identify single nucleotide variants or copy number variations of importance for DD and SI. ResultsIndividual 1 had primary ciliary dyskinesia (PCD), a rare, autosomal recessive disorder with oto-sino-pulmonary phenotype and SI. We identified two rare nonsynonymous variants in the dynein axonemal heavy chain 5 gene (DNAH5): c.7502G>C;p.(R2501P), a previously reported variant predicted to be damaging and c.12043T>G;p.(Y4015D), a novel variant predicted to be damaging. Ultrastructural analysis of the cilia revealed a lack of outer dynein arms and normal inner dynein arms. MRI of the brain revealed no significant abnormalities. Individual 2 had non-syndromic SI and DD. In individual 2, one rare variant (c.9110A>G;p.(H3037R)) in the dynein axonemal heavy chain 11 gene (DNAH11), coding for another component of the outer dynein arm, was identified. ConclusionsWe identified the likely genetic cause of SI and PCD in one individual, and a possibly significant heterozygosity in the other, both involving dynein genes. Given the present evidence, it is unclear if the identified variants also predispose to DD, but further studies into the association are warranted.

genetics↗