A. Ercüment Çiçek

This paper introduces REFCON, a deep-learning model for reference-free copy number inference in single-cell tumor transcriptomes, showing improved accuracy and robustness compared to existing methods.

The paper introduces RNAGEN, a GAN-based model to generate synthetic RNA sequences that bind to target proteins, validated with in vitro experiments for SOX2.

DeepSynBa predicts complete dose-response matrices for drug combinations, outperforming existing methods and enabling more precise and actionable drug synergy analysis.

This paper shows that using one-hot encodings for drugs and cell lines in synergy prediction models can perform as well as or better than using molecular features, suggesting models rely more on identifiers than features.

RNA-X is a foundation model for designing RNA sequences that bind to multiple targets, including proteins, RNA, and DNA, and it outperforms existing methods in binding prediction and design.

ExactCN improves copy number variation estimation from whole-exome sequencing data using deep learning, achieving better accuracy than existing methods.

This paper introduces a reinforcement learning-based method to dynamically protect genomic data sharing beacons against membership inference attacks by adapting strategies in real time.

CHALLENGER improves CNV detection in challenging genomic regions using short-read WGS data, achieving higher F1-scores and capturing CNVs typically requiring long reads.

RNAtranslator generates RNA sequences that bind to specific proteins by treating RNA design as a sequence-to-sequence translation task, producing novel and high-affinity sequences.

LYCEUM is a machine learning-based CNV caller for ancient DNA that improves detection accuracy in low-coverage samples by pre-training on high-quality data and fine-tuning on limited aDNA samples.

This paper identifies a privacy vulnerability in genomic data-sharing beacons, demonstrating genome reconstruction using summary statistics and SNP correlations, achieving a higher F1-score than baseline methods.

UTRGAN generates 5' UTR sequences optimized for higher translation efficiency and gene expression, outperforming existing UTRs in multiple metrics.

This paper introduces PDSP, a deep learning framework that uses patient-specific single-drug response data to predict personalized drug synergy, improving accuracy by 27% in leukemia patients.

ECOLE is a deep learning model for detecting copy number variants in whole exome sequencing data, achieving improved precision and recall over existing methods.

PiDeeL improves survival analysis and pathological classification of gliomas by integrating pathway information into a deep learning model, achieving better performance metrics than existing methods.

This paper introduces Macarons, a fast algorithm for selecting complementary genetic variants to predict quantitative phenotypes, achieving similar or better performance than existing methods with significantly reduced runtime.

This paper introduces ATAC-DoubletDetector, a read count-based method to detect multiplets in snATAC-seq data and annotate their cellular origins, achieving 85% accuracy in annotation.

MatchMaker predicts drug synergy using chemical structures and gene expression, improving performance over existing methods on a large drug combination dataset.

This paper introduces RoKAI, a network-based framework that improves kinase activity inference by integrating functional information, showing enhanced accuracy and robustness in identifying dysregulated kinases in diseases.

This paper improves whole cell biosensors by integrating neural networks, achieving faster response times and higher accuracy in detecting analyte concentrations.

The paper introduces an epistasis test prioritization algorithm that improves precision and reduces computational load in GWAS by selecting diverse and complementary SNPs.

This study shows that including family members in genomic data-sharing beacons reduces re-identification attack success by obscuring variant information, but also causes some utility loss.

This paper introduces DeepND, a multitask learning algorithm for gene risk prioritization in comorbid neurodevelopmental disorders, identifying key brain regions and developmental windows associated with ASD and ID.

DECoNT improves CNV detection accuracy on WES data by learning from matched WES and WGS data, tripling duplication precision and doubling deletion precision.

The paper uses whole-genome and RNA sequencing from the human prefrontal cortex to identify genetic variants and transcriptomic changes associated with neurodevelopment and neuropsychiatric disorders.

This study introduces DeepSide, a deep learning framework for drug side effect prediction, showing improved performance using drug chemical structure and gene expression data.

SPADIS selects diverse SNPs for GWAS to improve predictive performance by prioritizing remote SNPs in SNP-SNP networks, outperforming SConES in regression and gene identification.

driverMAPS improves cancer driver gene detection by modeling positive selection and mutation patterns, identifying 159 new potential driver genes including RNA methyltransferases.

This paper introduces a spatio-temporal gene discovery algorithm for ASD that uses evolving gene coexpression networks to identify risk clusters with improved prediction power.

Hercules corrects long read errors using a profile HMM-based machine learning approach, achieving high mapping rates and accuracy when short reads cover most of the long reads.