Performance comparison of NAVF-Bio and 11 SOTA models in the ALK mutation prediction task based on the SXHCSU dataset.

The persistent high burden of lung cancer in China highlights a critical demand for the identification of new therapeutic targets and intervention approaches. Our initial integrative analysis of metabolomic and transcriptomic data revealed a previously uncharacterized tumor-suppressive mechanism mediated by CCDC6 in lung adenocarcinoma. We discovered an interaction between CCDC6 and ASS1, concomitant with marked reductions in citrulline and aspartate within tumor tissues. compared to the adjacent normal tissues. Additionally, co-stimulation with citrulline and aspartate induces ASS1 localization in the mitochondria ASS1 localized in the mitochondria, but the underlying mechanism remains unclear. This study aimed to delineate the dual tumor-suppressive actions of ASS1: firstly, through the recruitment of the deubiquitinase OTUD7A to remove ubiquitin chains from MFN1/2 and OPA1, thereby stabilizing the mitochondrial fusion machinery and inducing hyperfused network formation; and secondly, via the CCDC6-ASS1 complex, which instigates mitochondrial reactive oxygen species accumulation, compromises ATP synthesis, and reduces mitochondrial membrane potential, consequently inducing a state of metabolic dormancy in tumor cells. This study elucidates the mechanism by which the ASS1-CCDC6 axis suppresses lung adenocarcinoma progression by remodeling mitochondrial dynamics and metabolic homeostasis,, thereby establishing a theoretical basis for mitochondria-targeted precision therapy.

The ablation experiment results of the influence of TME and multi-scale features on NAVF-Bio prediction performance were statistically analyzed using the five-fold cross-validation method.

Integrin activation is an indispensable step for various integrin-mediated biological functions. Kindlin-2 is known to coactivate integrins with Talin; however, molecules that restrict integrin activation are elusive. Here, we demonstrate that the E3 ubiquitin ligase Smurf1 controls the amount of Kindlin-2 protein in cells and hinders integrin activation. Smurf1 interacts with and promotes Kindlin-2 ubiquitination and degradation. Smurf1 selectively mediates degradation of Kindlin-2 but not Talin, leading to inhibition of αIIbβ3 integrin activation in Chinese hamster ovary cells and β1 integrin activation in fibroblasts. Enhanced activation of β1 integrin was found in Smurf1-knockout mouse embryonic fibroblasts, which correlates with an increase in Kindlin-2 protein levels. Similarly, a reciprocal relationship between Smurf1 and Kindlin-2 protein levels is found in tissues from colon cancer patients, suggesting that Smurf1 mediates Kindlin-2 degradation in vivo. Collectively, we demonstrate that Smurf1 acts as a brake for integrin activation by controlling Kindlin-2 protein levels, a new mechanism that permits precise modulation of integrin-mediated cellular functions.

Performance comparison of NAVF-Bio and 11 SOTA models on the TP53 exon prediction task based on the SXHCSU dataset.

Performance comparison of NAVF-Bio and 11 SOTA models on the TMB status prediction task based on the SXHCSU dataset.

Quantitative TME metrics and correlation analysis of the TP53&EGFR, TP53&KRAS, and TP53&ALK groups compared to the ALK mutation group, with statistical significance determined by T-test.

Deep learning (DL) has the potential to enable the prediction of gene mutations directly from routine histopathology slides in lung cancer. However, existing approaches have largely been limited to mutation-level prediction and have not achieved precise identification of driver mutation subtypes nor exonic variants, constraining the translation of DL into targeted therapy. In this study, we assembled a large multicenter dataset of paired pathology images and next-generation sequencing from 2,573 patients with lung cancer from four hospitals in China. The development of NAVF-Bio, an adaptive multiview feature fusion framework based on multiple instance learning, enabled the integration of tumor microenvironment (TME) features from whole-slide images (WSI) to predict driver mutations and tumor mutational burden (TMB). Benchmarking against 11 state-of-the-art DL methods indicated that NAVF-Bio consistently outperformed existing models in predicting driver mutations (TP53, EGFR, KRAS, ALK) and TMB status, achieving clinically relevant performance in external multicenter validation. Notably, NAVF-Bio accurately predicted the mutated driver gene exons across centers, whereas interpretability analyses using WSI visualization and TME quantification further demonstrated the ability of NAVF-Bio to elucidate pathologically relevant tumor features. Finally, a multigene mutation prediction platform for lung cancer was generated to facilitate the screening of driver gene mutations. Overall, NAVF-Bio mimics the workflow of pathologists when examining slides by observing multiscale features of WSIs and TME characteristics to predict driver gene mutations in lung cancer, which could guide the selection of targeted therapies for patients. SIGNIFICANCE:The NAVF-Bio framework accurately predicts key driver gene mutations in lung cancer from routine pathology slides, offering opportunities for the application of AI in precision oncology.

Detailed information on some gene mutations of patients in the SXHCSU dataset.

Interpretability analysis of WSI for NAVF-Bio prediction of KRAS mutations.

Prov-GigaPath was compared with state-of-the-art pathology-based models using AUC on 10 pathomic and cancer subtyping tasks.

Background:Although immune checkpoint blockade (ICB) therapy has improved clinical outcomes for some patients with lung adenocarcinoma (LUAD), only a subset of cases can achieve durable benefits, and primary resistance is often associated with an immune-excluded tumor microenvironment (TME). Existing studies indicate that cancer-associated fibroblasts (CAFs) are increasingly recognized as important participants in extracellular matrix (ECM) remodeling and stroma-immune crosstalk. However, in LUAD, it is still unclear which CAFs-related programs are involved in immune exclusion, particularly in relation to their spatial interactions with myeloid cell states. Methods:We integrated eight publicly available LUAD single-cell RNA-seq cohorts (164 samples; 471,501 cells) using Harmony and annotated major lineages and stromal subsets. CAFs were reclustered to resolve subtype heterogeneity, followed by pathway activity scoring, weighted gene coexpression network analysis (WGCNA), pseudotime trajectory inference (Slingshot), and regulon analysis. Bulk TCGA-LUAD data were used for immune-exclusion correlation and survival analyses. Spatial transcriptomics was applied for in situ validation, and ligand-receptor analysis together with NicheNet was used to prioritize CAFs-derived signaling interactions and downstream targets. Results:The integrated atlas identified 10 major cell lineages and revealed tumor-associated expansion of stromal and myeloid compartments. We further resolved six CAFs subtypes with distinct molecular and functional features. Tumor-enriched CAFs subsets showed stronger activation of ECM remodeling, focal adhesion, TGF-β signaling, hypoxia-related pathways, and inflammatory programs. Bulk-level analyses in TCGA-LUAD demonstrated that CAFs-related signatures were associated with a T-cell exclusion index and increased expression of the immune checkpoint-related molecule CD276 (B7-H3). Spatial transcriptomic mapping further showed that selected CAFs signatures were enriched in hypoxic, nonepithelial regions and colocalized with myeloid-associated signals, supporting the presence of a spatially constrained fibro-myeloid niche. Cell-cell communication analysis revealed extensive ligand-receptor interactions between CAFs subtypes and myeloid populations, whereas survival and immunotherapy cohort analyses showed that specific CAFs-related programs were associated with worse clinical outcomes and less favorable treatment responses. Conclusions:Multicohort single-cell integration and spatial validation define a CAFs-centered, immune-excluded niche in LUAD characterized by coordinated stromal and myeloid programs. These findings improve current understanding of CAFs heterogeneity and fibroblast-myeloid coupling in LUAD and provide a framework for future strategies aimed at targeting stromal barriers and remodeling the immune microenvironment to enhance ICB responsiveness.

Performance comparison of NAVF-Bio and 11 SOTA models on the TP53 mutation prediction task based on the SXHCSU dataset.

Interpretability analysis of WSI in mutation group patients with high and low TMB status predicted by NAVF-Bio.

Performance comparison of NAVF-Bio and 11 SOTA models on the EGFR exon prediction task based on the SXHCSU dataset.

Integrating different data modalities of cancer patients can significantly improve predictive performance for patient survival. However, most existing methods fail to efficiently leverage the rich semantic features inherent in highly heterogeneous multimodal data. When collecting multimodal data and extracting features, some intramodal data may be missing, which can introduce noise into the multimodal dataset. To address these challenges, this paper introduces a new end-to-end framework, FORESEE, for robustly predicting patient survival by mining multimodal information. Specifically, the cross-fusion transformer effectively aligns features at the cellular, tissue, and tumor heterogeneity levels, facilitating prognosis predictions via a cross-scale feature cross-fusion method. This approach enhances the ability of pathological image feature representation. Second, the hybrid attention encoder (HAE) incorporates a denoising contextual attention module to obtain the contextual relationship features and local detail features of molecular data. The channel attention module in the HAE is designed to obtain the global features of the molecular data. Furthermore, to address the issue of missing information within modalities, we propose the use of an asymmetrically triplet masked autoencoder. Extensive experiments demonstrate the superiority of our method over state-of-the-art methods across four benchmark datasets in both complete and missing data scenarios. Our codes are available at https://github.com/panliangrui/FORESEE.

Accurate intraoperative and postoperative diagnosis of spread through air spaces (STAS) is essential for guiding surgical decisions and postoperative management in lung cancer. However, histopathological assessment is labor-intensive and is prone to missed or incorrect diagnoses. We propose a Diffusion Attention Expert Model (DAEM) to detect STAS in frozen sections (FSs) and paraffin sections (PSs). Its diffusion attention expert module leverages full attention aggregation to learn multi-scale features from histopathological images, while a dual-branch architecture strengthens multi-scale feature representation. On an internal dataset, DAEM achieves AUCs of 0.8946 for FSs and 0.9112 for PSs. Validation on external multi-center datasets from eight institutions demonstrates strong generalizability and interpretability. Using tumor microenvironment (TME) features in PSs, we further enable semi-automatic measurement of STAS location and its distance from the primary tumor. Several quantitative TME metrics are identified as potential biomarkers for STAS, including micropapillary-type STAS. Overall, DAEM offers a clinically actionable framework for STAS assessment by enabling accurate and interpretable detection on FSs and PSs, supporting postoperative risk stratification through quantitative TME-based analysis.

Performance comparison of NAVF-Bio and 11 SOTA models on the KRAS exon prediction task based on the SXHCSU dataset.

For patients with hematologic malignancies, novel therapeutic strategies offer the potential to achieve a complete clinical response and long-term survival. However, declining fertility has become a significant concern, impacting long-term quality of life. Conventional high-dose chemotherapy and radiotherapy are known to reduce fertility or cause sterility. Moreover, limited clinical data are available on the effects of newer therapies, such as targeted treatments and chimeric antigen receptor (CAR)-T cell therapy, on fertility. Additionally, there is no standard method for preserving fertility in these patients. Male patients can opt for sperm cryopreservation, whereas female patients may preserve fertility through embryo, oocyte, or ovarian tissue cryopreservation. However, preserving fertility in prepubescent patients remains particularly challenging. Therefore, hematologists must educate patients about the potential gonadal toxicity of cancer treatments and offer the most appropriate fertility preservation options.