Shanfeng Zhu

SCGBinner improves long-read metagenomic binning by using single-copy gene guided contrastive learning, outperforming existing methods on real-world datasets and recovering novel microbial species.

This paper organizes a community challenge to develop automated virus taxonomy classification tools and evaluates their performance on a combined dataset of expert-curated and metagenomic sequences.

VACmap improves detection of complex genomic rearrangements like inversions and duplications, which are critical for understanding diseases such as cardiovascular disease and Parkinson's disease.

PLMSearch uses a protein language model to perform fast and accurate homologous protein sequence search, improving sensitivity over traditional methods while matching structure search performance.

CLAlign improves protein alignment by using contrastive learning to generate structure-aware embeddings, outperforming both unsupervised and sequence-based methods.

NetGO 3.0 integrates a protein language model with NetGO 2.0 to improve functional annotations by leveraging unannotated proteins.

MetaBinner improves metagenomic binning by integrating multiple features and SCG information, outperforming existing methods on simulated and real datasets.

Phen2Disease prioritizes diseases and genes using phenotype similarity scores calculated from HPO terms, outperforming existing methods in cohorts with fewer HPO terms.

DeepMHCII improves MHC II-peptide binding affinity prediction by incorporating binding core-aware convolution layers, outperforming existing methods across multiple evaluation settings.

The paper evaluates metagenomic software tools using realistic datasets, highlighting improvements in assembly and taxon profiling while identifying remaining challenges in virus/archaea detection and reproducibility.

BERTMeSH improves MeSH indexing by using deep learning and full-text analysis, achieving higher accuracy and faster performance than existing methods.

NetGO improves large-scale protein function prediction by integrating sequence and network information, outperforming existing methods on CAFA benchmark data.

GOLabeler improves sequence-based protein function prediction by integrating multiple classifiers in a learning-to-rank framework, showing better performance than existing methods.