Luonan Chen

ProtMonomer improves protein structure prediction accuracy and efficiency by combining models trained on diverse evolutionary information levels, outperforming AlphaFold2 and AlphaFold3 on challenging targets.

The paper uses CERAMIC to trace liver tumor initiation, identifying Hep_Bi-zonal cells as the origin of HCC and ICC and revealing peroxisomal metabolism and macrophage niches as key factors.

NUMonomer predicts nucleic acid structures directly from sequence, outperforming existing methods without auxiliary inputs.

The paper identifies new genes involved in the evolution of parasperm classes in Drosophila, showing their role in sperm divergence and altruistic traits.

The paper tracks five-year longitudinal data of children learning abacus-based mental calculation to show how sustained practice reconfigures whole-brain states and enables complex skill acquisition.

This paper introduces InvDNA, a deep learning method for designing single-stranded DNA sequences from backbone coordinates, achieving better sequence recovery and structural accuracy than existing methods.

GeoBridge transforms sparse scRNA-seq data into a continuous landscape for reconstructing and navigating cellular state transitions.

This study benchmarks zero-shot single-cell foundation model embeddings for cellular dynamics reconstruction and finds they underperform a traditional HVG baseline in capturing complex temporal patterns.

VAETracer reconstructs cellular lineages and infers generational states from scRNA-seq data by analyzing mutation profiles, improving accuracy over noisy mutation signals.

The paper introduces MEMORY, a method to identify GEARs associated with cancer prognosis, finding mitosis-related genes in LUAD and immunity-related genes in BRCA.

DNFE detects tipping points in biological processes by transforming omics data into directed networks, showing robustness and identifying critical states and regulatory relationships.

BrainBeacon is a cross-species foundation model for whole-brain spatial transcriptomics that enables cross-species alignment and uncovers spatial regulatory mechanisms of aging.

ODFormer predicts personalized drug responses in pancreatic cancer by simulating patient-derived organoids using transcriptomic and mutational data, outperforming existing methods and improving clinical outcomes.

This paper introduces CarHE, a method that predicts spatial transcriptomics data from H&E images using contrastive learning, achieving high accuracy and enabling cost-effective analysis.

UniCure predicts personalized cancer therapy response by integrating omics and chemical models, achieving high accuracy in drug response and combination effects across diverse data.

This study introduces a causal inference algorithm based on cross-validated predictability, demonstrating high accuracy and robustness in identifying causal relationships and driver genes in liver cancer.

This study identifies statins as effective in targeting chemoresistant pancreatic cancer by inhibiting cholesterol metabolism and protein glycosylation, showing clinical benefit in a phase 2 trial.

This study identifies PIK3CA mutations as drivers of cachexia in EGFR-mutant lung cancer, showing that osimertinib can mitigate cachexia but chemotherapy fails, with aspirin offering a potential treatment.

CauFinder identifies and steers cell-state and phenotype transitions by causal disentanglement learning, outperforming existing methods in simulated and real-world scenarios.

STAMarker identifies spatial domain-specific variable genes using saliency maps in deep learning, outperforming existing methods on sparse spatial transcriptomic data.

This paper introduces mtANN, a method for cell-type annotation that identifies unseen cell types using multiple references, showing improved performance on benchmark and COVID-19 datasets.

The paper identifies an intermediate basal cell population (Basal-B) in the prostate that shows lineage plasticity and contributes to both homeostasis and cancer initiation.

LANTSA improves analysis of scRNA-seq and spatial transcriptomics data by addressing cross-platform and large-volume challenges through a novel subspace model and dimensionality reduction technique.

The paper introduces CCSN, a method to analyze scRNA-seq data by constructing conditional cell-specific networks to improve gene association analysis and cell clustering.

The paper investigates how intermediate states in EMT affect cancer metastasis by using a hidden Markov model to show that more intermediate states accelerate the process.

This paper introduces CPGD, a novel algorithm for identifying personalized driver genes and combinational drugs for individual cancer patients, showing improved performance over existing methods.

This paper introduces a nonlinear network control model to identify personalized driver genes in cancer by analyzing patient-specific molecular networks from TCGA.

The paper compares single-cell potency measures using 9 experiments and 6,600 RNA-Seq profiles, finding that graph entropy of a Markov Chain on a signaling network better estimates potency than other methods.

This paper introduces a sample-specific network method to construct individual-specific disease networks from single-sample expression data, identifying disease modules and driver genes without sequencing.