Biology subjects

Radwan, M.

Publications and source records attributed to Radwan, M..

6 recordsLinked to original sources

bioRxiv2026

Frequency-Resolved Connectivity Graphs for EEG-Based Classification of Alzheimer and Frontotemporal Dementia

In this study, we build EEG based graph model that uses frequency-resolved connectivity features, with Coherence, phase locking value (PLV) and phase lag index (PLI) in five frequency bands as multi-dimensional edge features. The model is used to classify Alzheimer disease (AD) and frontotemporal dementia (FTD) against healthy controls (CN). This paper extends our earlier work on FTD with AD vs CN as the main new task, a feature ablation for it, and a joint analysis of both tasks. With leave-one-subject-out (LOSO) validation, the model reaches an accuracy of 81.13\% for AD vs CN and 76.06\% for FTD vs CN, which is competitive with the baselines reported with the dataset with far fewer parameters (159 and 282 thousand versus 170.5 million). Under the random cross-validation setting used for the DICE-net model, our model has a comparable accuracy for FTD and a lower accuracy for AD. Node and edge saliency maps highlight occipital and temporal electrodes, the Alpha band for node features and the Theta band for edge features, which is consistent with earlier EEG studies. In an ablation on the AD task, using all 15 edge features performed better than using a single connectivity measure or edge features averaged over the bands.

neuroscience↗

bioRxiv2026

EEG Pathology Detection via Clinical Concept Bottleneck Models

Deep learning models for electroencephalography (EEG) pathology detection achieve competitive accuracy but offer limited interpretability. In this paper, we introduce concept bottleneck models (CBMs) for EEG pathology detection which use clinical EEG biomarkers computed analytically from the signal as a bottleneck to be used for prediction. Those concepts include band power, spectral ratios, alpha peak frequency and amplitude, hemispheric asymmetry and phase locking values. With the backbone based on shallow convolutional neural network (CNN), we compare a black box baseline with two variants of CBMs. The hybrid variant is a classifier receives a residual projection of the backbone features. The pure variant is a CBM where the classifier receives only the predicted concepts. Models are evaluated on abnormality detection and dementia classification followed by concept prediction quality and leakage analysis. Both CBMs approach the baseline on TUAB (82.7-82.9% vs 83.6%) and are 2-5 percentage points lower on CAUEEG (57.9-60.5% vs 62.7%). The pure CBM exhibits concept fidelity concern where the predicted concepts deviate from their ground truth values. Replacing them with their true values causes severe cost on accuracy by 26 and 22 points on TUAB and CAUEEG. The concept fidelity concern is mitigated by increasing the weight of the concept loss to match the ground truth concepts. the mitigation comes with small accuracy cost on TUAB but a relatively larger cost on CAUEEG. The code of the experiments and full logs are available at https://www.github.com/mhmdrdwn/cbm.

neuroscience↗

bioRxiv2025

Frequency bands EEG Biomarkers for Dementia using Graph Neural Networks

We introduce a simple and interpretable model for classification of electroencephalography (EEG) signals. Our focus essentially is on using deep learning to study how connectivity patterns that are integrated to classify the EEG signals and highlight the important discriminative features used by the model in predictions. In this study, we utilize the connectivity features across different frequency bands in multi edge Graph Neural Networks (GNN) and showed that edge features are complimentary. We use a simple GNN model to predict Frontotemporal Dementia (FTD) in EEG. Our model is capable of achieving average accuracy of approximately 76% using Leave-One-Subject-Out-subject for FTD predictions which are better than the baselines and comparable to State of the arts models. In this article, we study the importance of the connectivity edges, nodes and frequency bands in the prediction of the model, focusing in explainable AI methods through saliency maps to interpret the model both locally and globally. The Saliency maps highlight the importance of Occipital and anterior temporal regions in the prediction of FTD. Furthermore, our results highlight the importance of Alpha and Theta bands in the prediction of FTD. Our observations align with previous research done using classical statistical methods. We argue that there are complimentary information in each each connectivity feature and frequency band brain networks. The impacts of each connectivity metrics on the prediction of the model are quantified to highlight the complimentary information in each connectivity measure.

neuroscience↗

bioRxiv2025

Dynamic Graphs Analysis of EEG

In this study, we investigate the use of temporal dynamics in brain connectivity for the classification of electroencephalography (EEG) signals using dynamic Graph Neural Networks (GNNs). Our methods are applied to several large-scale EEG datasets focused on abnormality and epilepsy detection. The implemented models demonstrate competitive performance on unseen test subjects across all three datasets, outperforming previous graph-based baselines in terms of accuracy and F1 score. We explore multiple architectures designed to capture temporal variations in graph-structured data, demonstrating their effectiveness in modeling dynamic brain activity. In addition to classification, we employ graph-theoretical metrics to analyze temporal changes in brain networks, such as network efficiency and node degree, across time windows of EEG recordings. The goal is to characterize differences between pathological and healthy groups at both the node and network levels. We particularly examine epilepsy and healthy subject groups to highlight differences in local network efficiency and node degrees, with statistical significance confirmed via F-tests.

neuroscience↗

bioRxiv2020

Inclusion bodies formed by polyglutamine and poly(glycine-alanine) are enriched with distinct proteomes but converge in proteins that are risk factors for disease and involved in protein degradation

Poly(glycine-alanine) (polyGA) is one of the dipolypeptides expressed in Motor Neuron Disease caused by C9ORF72 mutations and accumulates as inclusion bodies in the brain of patients. Superficially these inclusions are similar to those formed by polyglutamine (polyQ) in Huntingtons disease and both have been reported to form an amyloid-like structure suggesting they might aggregate via similar mechanisms to confer cellular dysfunction similarly. Here we investigated which endogenous proteins were enriched in these inclusions and whether aggregation-prone lengths of polyQ (Q97), in context of Huntingtin exon 1, shared similar patterns to aggregation-prone lengths of polyGA (101GA). When co-expressed in the same cell, polyGA101 and HttQ97 inclusions adopted distinct phases with no overlap suggesting different endogenous proteins would be enriched. Proteomic analyses indeed yielded distinct sets of endogenous proteins recruited into the inclusion types. The proteosome, microtubules, TriC chaperones, and translational machinery were enriched in polyGA aggregates, whereas Dnaj chaperones, nuclear envelope and RNA splicing proteins were enriched in polyQ aggregates. Both structures revealed a synergy of degradation machinery including proteins in the polyQ aggregates that are risk factors for other neurodegenerative diseases involving protein aggregation when mutated, which suggests a convergence point in the pathomechanisms of these diseases.

biochemistry↗

bioRxiv2019

Arginine valency in C9ORF72 dipolypeptides mediates promiscuous proteome binding that stalls ribosomes, disable actin cytoskeleton assembly and impairs arginine methylation of endogenous proteins

C9ORF72-associated Motor Neuron Disease patients feature abnormal expression of 5 dipeptide repeat (DPR) polymers. Here we used quantitative proteomics in a Neuro2a cell model to demonstrate that the valency of Arg in the most toxic DPRS, PR and GR, drives promiscuous binding to the proteome, compared to a relative sparse binding of the more inert AP and GA. Notable targets included ribosomal proteins, translation initiation factors and translation elongation factors. PR and GR comprising more than 10 repeats robustly stalled the ribosome suggesting high-valency Arg electrostatically jams the ribosome exit tunnel during synthesis. Poly-GR also bound to arginine methylases and induced hypomethylation of endogenous proteins, with a profound destabilization of the actin cytoskeleton. Our findings point to arginine in GR and PR polymers as multivalent toxins to translation as well as arginine methylation with concomitant downstream effects on widespread biological processes including ribosome biogenesis, mRNA splicing and cytoskeleton assembly.\n\nSIGNIFICANCE STATEMENTThe major genetic cause of MND are mutations in an intron of the C9ORF72 gene that lead to the expansion in the length of a hexanucleotide repeat sequence, and subsequent non-AUG mediated translation of the intron into 5 different DPRs. The two DPRs containing Arg are potently toxic in animal and cell models. Our research shows that the valency of Arg mediates widespread proteome binding especially affecting machinery involved in Arg-methylation, cytoskeleton and translation. We suggest the mechanisms for toxicity are multipronged and involve electrostatic jamming of ribosomes during translation, acting as substrate mimetics for arginine methylase activity that renders the endogenous proteome hypomethylated and impairing actin cytoskeleton assembly. These mechanisms explain pathologic signatures previous reported in human brain pathology.

biochemistry↗