Olga L. Kardymon

The paper benchmarks gene annotation methods, showing fine-tuned DNA language models outperform existing tools and generalize across species and annotate non-coding transcripts.

GeomMotif evaluates protein generation models' ability to preserve arbitrary structural fragments, revealing differences in performance based on structural and physicochemical context.

GENA-LM introduces transformer-based DNA language models for long sequences, capable of handling up to 36,000 base pairs and supporting multispecies and taxon-specific applications.

This paper systematically evaluates metrics for protein generative models, identifying challenges and offering recommendations for robust evaluation.

GENA-Web is a web platform using DNA language models to annotate genomic elements, including promoter activity, splice sites, chromatin features, and enhancer activity in Drosophila.

The paper introduces DeepCT, a deep learning model that predicts cell type-specific effects of noncoding variants by learning from sparse epigenetic data.

PROSTATA predicts protein stability changes from point mutations using a transformer-based model, outperforming existing neural network approaches.

SEMA predicts conformational B-cell epitopes using transfer learning on antigen sequences and structures, achieving ROC AUC of 0.76 on an independent test set.