Accurate gene regulatory network (GRN) inference from single-cell RNA sequencing (scRNA-seq) data is challenged by high sparsity, noise, and confounders, causing confounding bias and ambiguous directions in traditional methods. To overcome this, we propose the Counterfactual Causality-aware Framework for Gene Regulatory Networks (CfCGRN). Building on counterfactual intervention, CfCGRN employs Conditional Variational Autoencoders (CVAEs) to generate perturbation-mimicking counterfactual samples. It incorporates a confounder-aware graph attention layer with learnable backdoor adjustments to explicitly eliminate confounding bias. An expression gradient-guided attention mask enforces directional information flow from highly to lowly expressed genes, enabling directed causal inference. Evaluations on DREAM5 and diverse scRNA-seq datasets demonstrate CfCGRN's significant improvements in GRN inference accuracy, directionality discrimination, and biological interpretability, offering a novel tool for deciphering key regulatory pathways.

Accurate prediction of protein-ligand affinity is a core component of drug discovery. Existing graph neural network methods mostly only update node features and ignore the explicit evolution of geometric coordinates, thereby failing to ensure full utilization of rotational/translational symmetry, which makes it difficult to characterize the complex coupling relationships between long-range and short-range interactions. To address this, we propose an Equivariant Hierarchical Graph Interaction Network (EHGIN), a hierarchical geometric interaction network with equivariance. Specifically, we designed an equivariant message passing layer that uses direction vector weights to achieve equivariant updates for coordinate rotation/translation. In addition, we used a 9dimensional radial basis function (RBF) to smoothly embed distances in the 0-6 Å range, balancing geometric resolution and computational efficiency. To validate the generalization ability of our model, we conducted tests on three external test sets and verified that our model performed optimally.

Cancer presents a complex global health challenge, with multi-omics data offering unprecedented insights into tumorigenesis. However, the high dimensionality and heterogeneity of these data demand advanced computational approaches. We propose a novel KAN-based feature fusion network for multi-omics survival analysis framework KANCurvNet, which integrates genomic, transcriptomic, epigenomic, and proteomic profiles while leveraging geometric curvature and patient-sample neighborhood relationships. Our method employs a variational autoencoder (VAE) for dimensionality reduction, a multi-scale factorized bilinear model (MC-FBM) for cross-omics representation, and graph convolutional networks (GCN) with Kolmogorov-Arnold Networks (KAN) to capture high-order patterns. A Cox proportional hazards model then predicts survival risk. Experiments across multiple cancer cohorts demonstrate superior performance over traditional methods while maintaining interpretability. This approach effectively addresses multi-omics heterogeneity and introduces a novel paradigm for deep learning in cancer research.

High-throughput sequencing technologies have facilitated a deeper exploration of prognostic biomarkers. While many deep learning (DL) methods primarily focus on feature extraction or employ simplistic fully connected layers within prognostic modules, the interpretability of DL-extracted features can be challenging. To address these challenges, we propose an interpretable cancer prognosis model called Cox-Sage. Specifically, we first propose an algorithm to construct a patient similarity graph from heterogeneous clinical data, and then extract protein-coding genes from the patient's gene expression data to embed them as features into the graph nodes. We utilize multilayer graph convolution to model proportional hazards pattern and introduce a mathematical method to clearly explain the meaning of our model's parameters. Based on this approach, we propose two metrics for measuring gene importance from different perspectives: mean hazard ratio and reciprocal of the mean hazard ratio. These metrics can be used to discover two types of important genes: genes whose low expression levels are associated with high cancer prognosis risk, and genes whose high expression levels are associated with high cancer prognosis risk. We conducted experiments on seven datasets from TCGA, and our model achieved superior prognostic performance compared with some state-of-the-art methods. As a primary research, we performed prognostic biomarker discovery on the LIHC (Liver Hepatocellular Carcinoma) dataset. Our code and dataset can be found at https://github.com/beeeginner/Cox-sage.

MOTIVATION: Constructing a phylogenetic tree requires calculating the evolutionary distance between samples or species via large-scale resequencing data, a process that is both time-consuming and computationally demanding. Striking the right balance between accuracy and efficiency is a significant challenge. RESULTS: To address this, we introduce a new algorithm, MIKE (MinHash-based k-mer algorithm). This algorithm is designed for the swift calculation of the Jaccard coefficient directly from raw sequencing reads and enables the construction of phylogenetic trees based on the resultant Jaccard coefficient. Simulation results highlight the superior speed of MIKE compared to existing state-of-the-art methods. We used MIKE to reconstruct a phylogenetic tree, incorporating 238 yeast, 303 Zea, 141 Ficus, 67 Oryza, and 43 Saccharum spontaneum samples. MIKE demonstrated accurate performance across varying evolutionary scales, reproductive modes, and ploidy levels, proving itself as a powerful tool for phylogenetic tree construction. AVAILABILITY AND IMPLEMENTATION: MIKE is publicly available on Github at https://github.com/Argonum-Clever2/mike.git.