Fengdi Zhao

BayesCNet improves inference of cell type-specific regulatory networks by leveraging cell type hierarchy, outperforming existing methods, especially for rare cell types.

Deep5hmC predicts genome-wide 5hmC modification using a multimodal deep learning model, outperforming unimodal models and existing methods in both qualitative and quantitative predictions.

This paper introduces scaDA, a statistical method for differential chromatin accessibility analysis in scATAC-seq data, showing improved power and FDR control compared to existing methods.

TIVAN-indel scores noncoding sindels for their regulatory potential using XGBoost and multi-omics features, achieving strong cross-tissue prediction performance.

The paper uses deep transfer learning to predict functional noncoding variants, outperforming existing methods on MPRA and GWAS datasets.

The paper develops multi-task deep autoencoders to predict Alzheimer's disease progression using longitudinal DNA methylation data from peripheral blood and shows improved performance over existing methods.