Qi Liu

AlphaCell simulates perturbation-induced cellular dynamics by addressing limitations in current models through three innovations: latent manifold rectification, biological reality reconstruction, and universal state transition.

CausCell improves explainability and controllability of single-cell data through causal disentanglement learning and counterfactual generation.

This paper benchmarks 40 single-cell multi-modal integration algorithms across paired, unpaired, and mosaic datasets to evaluate usability, accuracy, and robustness.

The paper introduces CHOOSER, a foundation model-based framework for discovering CRISPR-Cas systems with self-processing pre-crRNA capability, identifying 11 novel homologs and validating one for functional use.

This study introduces STAMP, a novel method for genetic perturbation outcome prediction that improves accuracy across three subtasks and provides a benchmarking framework for evaluating models.

PerturBase is a database that compiles and visualizes single-cell perturbation data from multiple studies, enabling analysis of genetic and chemical perturbations across millions of cells.

This study maps tumor evolution and microenvironment dynamics in colorectal cancer using spatial multi-omic data, identifying key genes associated with immune exclusion and prognosis.

The paper clarifies that improper negative data sampling strategies can bias TCR-antigen prediction models like PanPep and advocates for consistent comparison methods.

UniTCR integrates T-cell receptor and transcriptome data using contrastive learning for diverse immunology tasks, showing superior performance on paired datasets.

This study introduces TCRBagger, a weakly-supervised framework for peptide-TCR binding prediction to improve neo-epitope identification in tumor immunotherapy.

The paper develops contamination-focused metrics and software to optimize scRNA-seq experiments by reducing ambient contamination through experimental adjustments.

scPrivacy enables privacy-preserving integration of single-cell RNA-seq datasets from multiple institutions for cell type identification.

The paper proposes two computational methods to predict and characterize cell-type-specific and shared TF binding sites using XGBoost and CNN, showing improved performance over existing methods.

Dr. Sim is a learning-based framework for inferring transcriptional similarity to improve phenotypic drug discovery, outperforming existing methods on drug annotation and repositioning tasks.

The paper introduces a deep learning framework called FCNsignal to predict TF-DNA binding signals at base-resolution, outperforming existing methods and demonstrating its ability to locate binding regions and predict motifs.

X-MOL uses large-scale pre-training on 1.1 billion molecules to improve molecular analysis tasks through fine-tuning, achieving state-of-the-art results in property prediction, reaction analysis, and molecule generation.

This paper introduces FL-QSAR, a federated learning framework for QSAR analysis, demonstrating its effectiveness in collaborative drug discovery while preserving privacy.

The paper introduces a dual-indexed library design for scRNA-seq to improve sequencing quality and throughput on high-throughput sequencers.

ProGeo-neo integrates genomic and proteomic data to improve neoantigen prediction by incorporating mass spectrometry validation and MHC binding prediction.

The paper re-analyzes WGS data from two studies to find that CRISPR off-target mutations include both SNVs and indels, challenging previous conclusions about their rarity.