Ke Yuan

This paper uses self-supervised learning to discover histomorphological phenotypes from mesothelioma biopsies, linking tissue patterns to molecular and clinical outcomes.

This paper uses self-supervised learning on abdominal MRI data to identify associations with 158 diseases, including those not directly related to abdominal anatomy, and shows improved performance over standard biomarkers for MASH.

This study evaluates how protein language models capture mutational processes and fitness landscapes, finding that explicit mutational accessibility improves predictions while historical biases degrade performance.

ExplainBind predicts protein-ligand binding likelihood and identifies key residues and non-covalent interactions without structural inputs, outperforming existing methods and enabling drug discovery insights.

This study identifies architectural subtypes of PDAC using morphospatial profiling of CAFs, showing improved prognostication and chemotherapy response prediction.

This paper uses self-supervised AI to identify under-appreciated stromal morphologies in lung adenocarcinoma linked to poor prognosis, independent of current clinical schemas.

This paper introduces a self-supervised learning framework to analyze multiplex immunofluorescence images, revealing novel prognostic spatial patterns in cancer tissues.

This paper uses self-supervised learning on 3446 H&E-stained mesothelioma slides to create a histomorphological atlas, achieving high predictive performance and revealing molecular insights.

This paper uses self-supervised learning to identify histomorphological patterns in colon cancer WSIs, revealing clinically relevant clusters associated with survival and treatment responses.

This paper introduces EvoMIL, a deep learning method that predicts virus-host associations using protein language models and multiple instance learning, achieving improved accuracy in both prokaryotic and eukaryotic hosts.

PLM-interact improves PPI prediction by jointly encoding protein pairs, achieving 16-28% AUPR gains across multiple species.

This paper shows that protein language models can assess mutation effects in SARS-CoV-2 spike protein without multiple sequence alignments, capturing evolutionary constraints and variant properties.

The paper identifies pan-cancer evolutionary signatures linked to clonal expansion and immune microenvironment changes using TCGA data.

The paper uses machine learning to identify mutational signatures in SARS-CoV-2 genomes and links them to host antiviral molecules, showing these drive most of the virus's evolutionary changes.

This study shows that pre-trained protein language models can predict mutation pathogenicity and patient survival in cancer, outperforming existing methods.

FrenchFISH improves DNA copy number quantification from FISH images by using Poisson models, outperforming manual and automated methods in ovarian cancer tissue sections.

Ccube estimates cancer cell fractions and subclonal composition from whole-genome sequencing data using a probabilistic framework and variational inference.