MOTIVATION: The advent of single-cell RNA sequencing (scRNA-seq) technology has allowed researchers to measure gene expression profiles at the single-cell level, providing valuable insights into cellular heterogeneity. However, due to the limitations of current sequencing platforms, scRNA-seq data often contain significant noise, particularly severe dropout events, which pose major challenges for subsequent analyses. RESULTS: In this study, we developed a new method called topology-aware contrastive learning (scTACL). This approach uses contrastive learning between a cell similarity graph and a cell embedding similarity graph, employing a zero-inflated negative binomial (ZINB) distribution to model the reconstructed data. This alignment helps the processed data better reflect true biological signals. It delivers superior results in key tasks such as data imputation, clustering, batch effect correction, and cell-cell interaction. Additionally, scTACL successfully identified two distinct subtypes of epithelial cells in lung adenocarcinoma tissues, further demonstrating its effectiveness and usefulness in complex biological settings. Notably, without relying on spatial location information, scTACL still effectively distinguished the epithelial and mesenchymal regions in the spatial transcriptome data of liver cancer and identified the COLLAGEN signaling pathway, which plays a crucial role in the epithelial-mesenchymal transition process through intercellular communication analysis.

Accurate survival risk stratification in epithelial ovarian cancer remains challenging because prognostic information is distributed across heterogeneous clinical, histopathological, radiological, and molecular scales, while modality availability is often incomplete across cohorts. We present a compact fuzzy-rule decision-level fusion framework centered on a primary clinical-histopathology survival model and extended through controlled modality-extension analyses. The primary model operates on calibrated unimodal risk scores and integrates fuzzy membership embedding, rule screening, and compact rule distillation to produce a frozen survival score for downstream use. On the clinical-histopathology complete-case subsets, the compact model achieved C-indices of 0.6771 in TCGA-OV and 0.6085 in the independent Memorial Sloan Kettering Cancer Center cohort, and yielded the strongest external discrimination among the evaluated two-modality late-fusion comparators. Paired bootstrap analysis showed significant gains over the clinical unimodal baseline and QMF, while the remaining pairwise comparisons were directionally favorable but not uniformly significant. CT radiomics, evaluated as an auxiliary modality under incomplete availability, provided only modest local refinement and did not redefine the primary model. In the matched molecular subset, transcriptomics provided substantial complementary value beyond the frozen primary score, whereas reverse incremental analysis showed that the primary model retained nonredundant prognostic information beyond the molecular score. Exploratory biological analyses linked the joint molecular extension score to attenuation of immune- and module-related programs and to enrichment of extracellular-matrix and migratory pathways in high-risk tumors. These findings support a compact, interpretable, and deployment-oriented decision-level fusion strategy for ovarian cancer survival modeling.

The development of spatial transcriptomics (ST) technologies has enabled researchers to better understand cells' spatial organization and functional heterogeneity within their native tissue context. Spatial domain identification plays a crucial role in ST data analysis. However, most existing spatial domain identification methods do not fully exploit spatial information, and often fail to adequately integrate both local and global features, resulting in suboptimal spatial domain identification. We propose SLGCA, a novel method based on cross-level graph contrastive learning to address these challenges. SLGCA adopts a dual-channel learning mechanism, combining local-level contrastive learning based on spatial neighborhood information and global information contrastive learning across views, thereby significantly enhancing the accuracy of spatial domain identification. SLGCA can integrate multiple tissue sections without needing pre-alignment or external tools, eliminating batch effects and accurately identifying spatial domains across multiple slices. Experimental results show that SLGCA significantly outperforms the benchmark methods in spatial domain identification accuracy on ST data generated by multiple techniques. Moreover, SLGCA enables accurate dissection of tumor heterogeneity in human breast cancer datasets and effectively uncovers the heterogeneous tumor microenvironment in liver cancer, revealing two distinct fibroblast subtypes.

Interactions between long non-coding RNAs (lncRNAs) and microRNAs (miRNAs) play an important role in the development of complex human diseases by collaboratively regulating gene transcription and expression. Therefore, identifying lncRNA-miRNA interactions (LMIs) is essential for diagnosing and treating complex human diseases. Because identifying LMIs with wet experiments is time-consuming and labor-intensive, some computational methods have been developed to infer LMIs. However, these approaches excel at utilizing single-modal information but struggle to integrate multimodal data from lncRNAs and miRNAs, which is essential for uncovering complex patterns in LMIs, ultimately limiting their performance. Therefore, this article proposes a novel multimodal contrastive representation learning model (MCRLMI) for LMI predictions. The model fully integrates multi-source similarity information and sequence encodings of lncRNAs and miRNAs. It leverages a graph convolutional network (GCN) and a Transformer to capture local neighborhood structural features and long-distance dependencies, respectively, enabling the collaborative modeling of structural and semantic information. Subsequently, to effectively integrate multimodal characteristics with encoded information, a multichannel attention mechanism and contrastive learning are introduced to fuse the extracted features. Finally, a Kolmogorov-Arnold Network (KAN) is trained with the optimized embeddings to predict LMIs. Extensive experiments show that the proposed MCRLMI consistently outperforms existing methods. Moreover, case studies further validate the potential of MCRLMI to identify novel LMIs in practical applications.

Transfer learning has been widely applied to drug sensitivity prediction based on single-cell RNA sequencing, leveraging knowledge from large datasets of cancer cell lines or other sources to improve the prediction of drug responses. However, previous studies require model fine-tuning for different patient single-cell datasets, limiting their ability to meet the clinical need for high-throughput rapid prediction. In this research, we introduce single-cell Adaptive Transfer and Distillation model (scATD), a transfer learning framework leveraging large language models for high-throughput drug sensitivity prediction. Based on different large language models (scFoundation and Geneformer) and transfer strategies, scATD includes three distinct sub-models: scATD-sf, scATD-gf, and scATD-sf-dist. scATD-sf and scATD-gf employs an important bidirectional style transfer to enable predictions for new patients without model parameter training. Additionally, scATD-sf-dist uses knowledge distillation from large models to enhance prediction performance, improve efficiency, and reduce resource requirements. Benchmarking across more diverse datasets demonstrates scATD's superior accuracy, generalization and efficiency. Besides, by rigorously selecting reference background samples for feature attribution algorithms, scATD also provides more meaningful insights into the relationship between gene expression and drug resistance mechanisms. Making scATD more interpretability for addressing critical challenges in precision oncology.

Automatic acne severity grading is crucial for the accurate diagnosis and effective treatment of skin diseases. However, the acne severity grading process is often ambiguous due to the similar appearance of acne with close severity, making it challenging to achieve reliable acne severity grading. Following the idea of fuzzy logic for handling uncertainty in decision-making, we transforms the acne severity grading task into a fuzzy label learning (FLL) problem, and propose a novel adjacency-aware fuzzy label learning (AFLL) framework to handle uncertainties in this task. The AFLL framework makes four significant contributions, each demonstrated to be highly effective in extensive experiments. First, we introduce a novel adjacency-aware decision sequence generation method that enhances sequence tree construction by reducing bias and improving discriminative power. Second, we present a consistency-guided decision sequence prediction method that mitigates error propagation in hierarchical decision-making through a novel selective masking decision strategy. Third, our proposed sequential conjoint distribution loss innovatively captures the differences for both high and low fuzzy memberships across the entire fuzzy label set while modeling the internal temporal order among different acne severity labels with a cumulative distribution, leading to substantial improvements in FLL. Fourth, to the best of our knowledge, AFLL is the first approach to explicitly address the challenge of distinguishing adjacent categories in acne severity grading tasks. Experimental results on the public ACNE04 dataset demonstrate that AFLL significantly outperforms existing methods, establishing a new state-of-the-art in acne severity grading.

Structural variation (SV) is an important form of genomic variation that influences gene function and expression by altering the structure of the genome. Although long-read data have been proven to better characterize SVs, SVs detected from noisy long-read data still include a considerable portion of false-positive calls. To accurately detect SVs in long-read data, we present SVDF, a method that employs a learning-based noise filtering strategy and an SV signature-adaptive clustering algorithm, for effectively reducing the likelihood of false-positive events. Benchmarking results from multiple orthogonal experiments demonstrate that, across different sequencing platforms and depths, SVDF achieves higher calling accuracy for each sample compared to several existing general SV calling tools. We believe that, with its meticulous and sensitive SV detection capability, SVDF can bring new opportunities and advancements to cutting-edge genomic research.

With the increasing volume of high-throughput sequencing data from a variety of omics techniques in the field of plant-pathogen interactions, sorting, retrieving, processing and visualizing biological information have become a great challenge. Within the explosion of data, machine learning offers powerful tools to process these complex omics data by various algorithms, such as Bayesian reasoning, support vector machine and random forest. Here, we introduce the basic frameworks of machine learning in dissecting plant-pathogen interactions and discuss the applications and advances of machine learning in plant-pathogen interactions from molecular to network biology, including the prediction of pathogen effectors, plant disease resistance protein monitoring and the discovery of protein-protein networks. The aim of this review is to provide a summary of advances in plant defense and pathogen infection and to indicate the important developments of machine learning in phytopathology.