Nilah M. Ioannidis

The paper introduces MeRN, a graph-guided variational autoencoder for learning single-cell metabolic states from transcriptomes, and identifies metabolic drivers in various biological contexts.

A sequence-only convolutional neural network and an annotation-based random forest predict human structural variant breakpoints from 400 bp windows (AUROC 0.903, 0.904); their ensemble reaches 0.916.

This study identifies 35.3 million structural variants across vertebrates, showing significant variation in their abundance, mechanisms, and functional impact across species.

GUANinE v1.1 evaluates supervised and unsupervised genomic models, showing each excels on specific tasks and identifying tradeoffs in model performance and resource use.

This paper introduces MethylSeqNet, a model that improves sequence-based gene regulation predictions by conditioning on CpG methylation, showing benefits for cell-type-specific regulation and allele-specific phenomena.

This study fine-tunes sequence-to-expression models on personal genome and transcriptome data, showing improved performance on seen genes but not on unseen genes.

PCLAI infers continuous genetic ancestry across the genome using a point cloud representation, enabling time-stratified geographic insights from ancient samples.

This paper introduces a stability-guided fine-mapping approach that complements residualization, showing it enhances functional variant identification in eQTL analysis.

Splice Ninja predicts tissue-specific splicing events using splicing factor expression, achieving accurate results on unseen tissues.

The paper introduces dimelo-toolkit, a Python package for analyzing multimodal long-read epigenetic data, demonstrating its use with a novel DiMeLo-seq protocol to study chromatin state in GM12878 cells.

The paper introduces a model that improves transcription factor binding prediction by integrating DNA sequence and structural features, enhancing interpretability and regulatory element identification.

The paper evaluates missense variant effect predictors using CAGI6 data, finding strong performance across settings and highlighting biases in some predictors.

This paper identifies genetic variants in tubule epithelial regulatory elements that explain most heritable differences in human kidney function using GWAS, scATAC-seq, and functional assays.

This paper introduces Enformation Theory, a framework to evaluate and interpret deep learning models in genomics using Enformer and Borzoi as case studies.

The paper evaluates deep learning models for predicting chromatin accessibility and finds reduced performance in cell type-specific regions, suggesting strategies to improve accuracy.

The paper introduces a framework for designing cell-type-specific promoters using model-based optimization, demonstrating improved specificity in blood cancer cell lines.

This study examines uncertainty in genomic sequence-to-activity model predictions by analyzing consistency across different sequence types and identifies areas where models perform poorly.

This study evaluates genomic deep learning models' ability to explain individual expression variation and finds limited performance.

GUANinE v1.0 provides benchmark datasets for evaluating genomic AI sequence-to-function models across functional genomics and evolutionary biology tasks.

This paper introduces a cross-protein transfer learning framework that improves disease variant prediction using deep mutational scanning data from few proteins, achieving better specificity than existing models.

This paper develops transfer learning strategies to model promoter-driven gene expression in understudied cell types, showing improvements in performance with pretraining and transfer.

This study quantifies how aging and genetics influence gene expression across 27 human tissues, finding tissue-specific differences in their contributions and identifying signatures of aging and selection.