CSGDA: A Cell State-Guided Graph Domain Adaptation Network for Single-Cell Drug Response Prediction

CAR T cell foundation model predicts immunotherapy response

Single-cell transcriptomics resolves CAR T-cell states, yet translating heterogeneous cellular signals into patient-level therapeutic response remains challenging. Existing studies primarily identify response-associated genes or cell populations through experimental and statistical analyses, but few predictive frameworks integrate gene-level structure with clinical outcomes. Here, we present gANCHOR, a T-cell foundation model built on a hierarchical hypergraph attention framework combining biologically informed representation learning with patient-level response prediction. By encoding gene-pathway relationships, gANCHOR learns pathway-aware cell embeddings that improve biological conservation and batch robustness. A cell-to-patient attention module then aggregates cellular information to infer therapeutic response. Across benchmark datasets, gANCHOR achieved the strongest overall performance in biological conservation and batch-correction assessments. In response prediction across 161 patients from five CAR T-cell studies, gANCHOR achieved an F1 score of 0.87, outperforming benchmarked single-cell foundation models. gANCHOR also identified reproducible response- and non-response-associated gene programs, providing interpretable biological insights into CAR T-cell efficacy and resistance.

Drug target prediction from perturbation transcriptomics via a biological function-guided hypergraph siamese network

Abstract Motivation Understanding how small molecules modulate cellular states remains a critical challenge in drug discovery. The advent of perturbation transcriptomics offers new avenues for elucidating drug-target interactions by capturing cellular transcriptional responses to perturbations. Results In this study, we propose BioHSNet, a biological function-guided hypergraph siamese network for inferring drug-target interactions from perturbation transcriptomics. BioHSNet utilizes hyperedge representations of functionally grouped gene expression to capture higher-order functional relationships, and integrates compound structural information into the model to bridge chemical structure and functional response. Experimental results demonstrate that BioHSNet outperforms other transcriptome-based methods on the Broad Institute’s L1000 datasets, particularly in cold start scenarios. The case study further demonstrates its practical utility for target prediction and drug screening. Availability and Implementation The source code is available at https://github.com/Zxinyizhang/BioHSNet.

Graph Neural Networks for Cancer Driver Gene Prediction: From Fundamentals to Applications.

A Biologically Informed Heterogeneous Graph Neural Network for Multi-Task Prediction of ncRNA–Metastasis–Cancer Interactions

Metastasis involves context-dependent molecular interactions in which non-coding RNAs, particularly miRNAs and circRNAs, play important regulatory roles. However, existing computational approaches generally do not jointly represent cancer type, metastatic event, and cancer-specific metastatic context. We developed a context-aware multi-task heterogeneous graph neural network (GNN) for predicting ncRNA associations with cancer types and metastatic events. The framework integrates multiple biological repositories into a heterogeneous graph representing ncRNAs, cancers, metastatic event types (METs), and cancer-specific metastatic instances (CSMIs). The model performs six link-prediction tasks using a hierarchical transformer-based encoder and multi-relational TuckER decoder. Across ten independently initialized runs evaluated on the RNA-group-disjoint held-out test set, the model achieved a global AUROC of 0.8801 ± 0.0118 and an F1 score of 0.8260 ± 0.0071. All three ablation variants yielded lower AUROC, with the largest reduction under independent task training. Case studies in pancreatic cancer, colorectal cancer, and hepatocellular carcinoma provided disease-level, event-level, and expression-based support, respectively, for top-ranked candidate associations. The framework enables context-specific prioritization of ncRNA–cancer–metastasis associations for experimental evaluation.

scGFormer: A Multi-Scale Graph-Transformer for Cell Type Annotation in Single-Cell RNA Sequencing.

Despite the rapid progress of single-cell RNA sequencing (scRNA-seq), accurate cell type annotation remains a major challenge. Existing approaches often struggle with sparse and heterogeneous expression profiles, insufficient genelevel modeling, and complications such as zero inflation and class imbalance. To address these issues, we propose scGFormer (Single-cell Multi-scale Graph Transformer), a unified framework that integrates: (i) Performer-based Global Attention (PGA) to capture long-range dependencies, (ii) Graph-based Local Attention (GLA) to model neighborhood structures, and (iii) a Squeeze-and-Excitation Gene Reweighting module (GeneSE) to enhance gene-level representations. Furthermore, scGFormer is equipped with a biology-guided adaptive contrastive learning strategy, which is designed to account for zero inflation, balance class distributions, and refine dynamic graphs during training, thereby facilitating robustness and adaptability. By explicitly modeling both global and local dependencies while strengthening gene-level representations, scGFormer achieves improved robustness and generalization. Extensive experiments across public datasets demonstrate that scGFormer achieves competitive or superior performance compared with state-of-theart methods, offering a robust solution for single-cell annotation across diverse datasets and species. Our code is publicly available at https://github.com/wuzi11/scGFormer.

Functionally Guided Graph Learning for Robust Cross-Patient Cell-Type Annotation in Single-Cell RNA Sequencing

Experiments on three cross-patient scRNA-seq data sets demonstrate that PathoGraph achieves stable annotation performance across 32 directed reference-to-query transfer tasks, showing competitive and stable performance compared with representative marker-based, correlation-based, and model-based annotation methods.