Dennis Kostka

This paper shows that low-rank approximations of DNA sequence-to-function models can achieve high performance while significantly reducing model size and computational cost.

The paper introduces MORALE, a domain adaptation framework for cross-species TF binding prediction, achieving state-of-the-art performance without adversarial training.

The paper introduces motifDiff, a tool for variant effect prediction using biophysical TF binding models, showing it effectively scores variants and provides insights into human accelerated regions.

This study shows that zebrafish retinal ganglion cells dedifferentiate after optic nerve injury, resembling immature RGCs, and identifies potential therapeutic targets for human RGC protection.

The paper analyzes PTCs in CHD patients to identify genes and pathways contributing to an oligogenic etiology of congenital heart disease.

The paper identifies trans-heterozygous digenic variants in laterality defect genes using exome/genome sequencing data from CHD cohorts, finding significant enrichment in laterality cases compared to controls.

The paper investigates the role of Tet2 and Tet3 in retinal development using scRNA-seq, identifying defects in cell fate specification and differentiation.

The paper develops a disease-specific variant prioritization method that improves the association of non-coding variants with diseases by aggregating tissue/cell-type specific scores.

aenmd annotates PTC-containing transcript-variant pairs for predicted NMD escape, identifying variants that may cause dominant-negative or gain-of-function effects.

The paper introduces tiSFM, an interpretable neural network for sequence-to-function learning, which outperforms existing models in predicting open chromatin and transcription factor activities.

The paper identifies metabolic regulation in left-right patterning using spatial transcriptome data, linking glycolysis and Bmp7 to heart and gut looping.

Vaeda annotates doublets in scRNA-seq data using a variational auto-encoder and Positive-Unlabeled learning, performing competitively and outperforming Python-based methods.

The paper introduces NIFA, a probabilistic factor analysis model for scRNA-seq data that disentangles discrete cell-type identity and continuous pathway activity, outperforming existing methods in biological interpretation.

The paper characterizes chromatin accessibility and microRNA expression in nephron progenitor cells during kidney development, identifying changes in regulatory elements and microRNAs associated with key biological processes.

This study uses single-cell RNA sequencing to identify differential cell cycle activity in nephron progenitors during mouse kidney development, highlighting Birc5 as a potential key molecule in nephron patterning.

The paper develops heart organoids and uses computational methods to model Ebstein's anomaly by analyzing single-cell RNA sequencing data from organoids with genetic variants.

The paper introduces scds, a tool for identifying doublets in scRNA-seq data, with two methods that perform competitively with existing approaches at lower computational cost.

The paper evaluates enhancer identification methods across four contexts, finding significant overlap but also substantial differences in genomic distribution, evolution, and function, and develops a database to address these challenges.

The paper shows that enhancer-promoter interaction prediction models can achieve competitive performance without capturing the underlying mechanism, highlighting issues with overfitting and cross-validation.

RAFSIL uses random forest to learn cell-cell similarities from scRNA-seq data, improving analysis by detecting technical variation and performing well across datasets.

The paper presents miRNA profiling data for mouse nephron progenitors, identifying differentially expressed miRNAs and novel miRNAs, and provides them as a public resource for kidney development studies.

The paper develops an attention-based neural network to predict enhancer-promoter interactions using DNA sequences and identifies key transcription factor binding sites.