Serdar Bozdag

DyGraphTrans models disease progression from EHRs using a dynamic graph framework, achieving strong predictive performance and interpretability.

The paper introduces CLEAR, a framework that aligns LLM embeddings with disease-specific knowledge graphs to improve drug repurposing, achieving state-of-the-art results on benchmark datasets.

MultiGEOmics integrates multi-omics data using graph-based methods to model cross-omics regulatory signals and handle missing data, showing strong performance across cancer and Alzheimer's datasets.

scAURA improves cell type clustering in scRNA-seq by integrating graph debiased contrastive learning, outperforming existing methods across multiple datasets.

This study evaluates how Alzheimer's disease diagnosis models perform across racial groups and finds that models trained on one group underperform on others, but fairness techniques can reduce bias.

Top-DTI predicts drug-target interactions by combining topological features and large language model embeddings, outperforming existing methods on benchmark datasets.

PDDRNet-MH predicts personalized drug responses in breast cancer by integrating multi-omics data and drug information, achieving high accuracy for several FDA-approved drugs.

The paper integrates spatial multi-omics data from diffuse gliomas to identify locoregional metabolomic regulators linked to glioma growth and treatment resistance.

This study identifies early miRNA expression changes in Arabidopsis infected with Fusarium graminearum, including novel miRNAs with potential for improving fungal resistance.

ProtFun predicts protein functions using a graph attention network with protein LLM embeddings, outperforming existing methods on benchmark datasets.

This study uses network biology methods to prioritize potential Alzheimer's disease-associated genes by integrating multimodal datasets and applying PhenoGeneRanker.

PVTAD classifies Alzheimer's disease and cognitively normal cases using a pyramid vision transformer on white matter MRI data, achieving 97.7% accuracy.

NRPreTo predicts Nuclear Receptor proteins and their subfamilies using enhanced features, achieving high performance and identifying 59 novel human NRs.

This paper introduces two deep learning models, PPAD and PPAD-AE, to predict Alzheimer's disease progression from mild cognitive impairment using EHR data, achieving better performance than baselines in F2 and sensitivity.

SUPREME predicts cancer subtypes using multiomics data and graph convolutions, outperforming existing methods and showing significant survival differences.

The paper evaluates cell lines across 22 cancer types to find those most similar to patient tumor samples using multi-omics data, immune cell presence, and cancer pathways, showing they can recapitulate drug responses.

Crinet infers genome-wide ceRNA networks by considering mRNAs, lncRNAs, and pseudogenes and using network deconvolution to exclude spurious interactions, showing enrichment in cancer-related genes.

NECo is a node embedding algorithm for multiplex heterogeneous networks that outperforms existing methods in classifying hypertension-related genes using rat disease models.

This paper develops an autoencoder and ANN model to estimate parity status of wild mosquitoes using NIR spectra, achieving high accuracy across multiple datasets.

CTDPathSim computes similarity between tumor samples and cell lines using multi-omics data to improve drug response prediction and guide personalized cancer therapies.

miRDriver identifies copy number-derived miRNA-gene regulatory interactions in cancer using integrated genomic data and outperforms existing methods.

PhenoGeneRanker prioritizes genes using a network propagation method on multi-omics data, showing enriched cold tolerance-related genes at the top.

This study examines the relationship between physiological arousal, self-reported anxiety, and facial expressions in autistic adolescents during a social skills intervention.