Chelsea J.‐T. Ju

JEDI predicts circular RNA from nucleotide sequences using attention mechanisms and cross-attention layers, outperforming existing methods and enabling backsplicing hotspot discovery.

Bio-JOIE learns joint representations of biological knowledge to predict protein interactions and cluster enzymes, outperforming existing methods in PPI type prediction.

This paper introduces MuPIPR, a deep learning framework that estimates mutation effects on protein-protein interactions by using contextualized amino acid representations and a Siamese residual recurrent convolutional encoder.

This paper evaluates neural network models for embedding Gene Ontology terms and introduces new models based on GCN, ELMo, and BERT, finding that BERT-based embeddings perform best for rare labels.

The paper analyzes 41 diseases and traits to show transposable elements contribute uniquely to disease heritability, with older elements and SINEs showing specific enrichment patterns.

Lasagna is an end-to-end framework for protein-protein interaction prediction using primary sequences, outperforming existing methods in binary prediction and showing promise in interaction type and binding affinity tasks.

The paper uses molecular QTL annotations to identify genetic variants associated with disease heritability, finding stronger enrichments using causal posterior probabilities than standard annotations.

TahcoRoll improves k-mer profiling by efficiently handling variable-length k-mers, outperforming existing tools in time and memory usage without disk space.

The paper develops a method to infer allelic heterogeneity in GWAS and eQTL datasets, identifying 4152 loci with strong evidence of AH across different traits.