Hereditas. 2025 Jul 7;162(1):121. doi: 10.1186/s41065-025-00487-4.

ABSTRACT

Accurate classification of cancer subtypes plays a pivotal role in advancing precision medicine. In this study, we introduce WMRCA + , a novel clustering approach based on a weighted majority rule that integrates multi-omics data and incorporates metabolic gene sets to robustly determine the optimal number of clusters for tumor subtype identification. WMRCA + evaluates clustering performance using ten internal metrics and offers comprehensive functionalities for data preprocessing and visualization. When applied to The Cancer Genome Atlas (TCGA) lung cancer dataset using lipid metabolism-related gene sets, WMRCA + outperformed widely used clustering algorithms-including iCluster, SNF, NMF, CC, and CNMF-achieving an AUC of 0.947. WMRCA + provides robust, interpretable, and biologically meaningful clustering results, offering a valuable tool for improving the accuracy of cancer subtype prediction. The WMRCA + R package is freely available at https://github.com/guojunliu7/WMRCA .

PMID:40624602 | PMC:PMC12235908 | DOI:10.1186/s41065-025-00487-4

Clin Transl Med. 2025 Jan;15(1):e70160. doi: 10.1002/ctm2.70160.

ABSTRACT

BACKGROUND: Plasma protein has gained prominence in the non-invasive predicting of lung cancer. We utilised Zeolite Zotero NaY-based plasma proteomics to investigate its potential for multiple event predicting, including lung cancer diagnosis (task #1), lymph node metastasis detection (task #2) and tumour‒node‒metastasis (TNM) staging (task #3).

METHODS: A total of 4703 plasma proteins were quantified from 241 participants based on a prospective cohort of 2757 participants. An additional 46 participants from external prospective cohort of 735 participants were used for validation. Feature selection was performed using differential expressed protein analysis, area under curve (AUC) evaluation and least absolute shrinkage and selection operator (LASSO) regression. Random forest was used for multitask model construction based on the key proteins. Feature importance was interpreted using Shapley additive explanations (SHAP) algorithm.

RESULTS: For task #1, 10 proteins panel showed an AUC of .87 (.77‒.97) in the external validation. After integrating clinical factors, a significant increase diagnostic accuracy was observed with AUC of .91 (.85‒.98). For task #2, nine proteins panel achieved an AUC of .88 (.80‒.96), integration model showed an increase diagnostic accuracy with AUC of .90 (.85‒.97). For task #3, 10 proteins panel showed an AUC of .88 (.74‒.96) for stage I, .92 (.84‒.97) for stage II, .88 (.76‒.96) for stage III and .99 (.98‒.99) for stage IV in the integration model.

CONCLUSIONS: This study comprehensively profiled the NaY-based plasma proteome biomarker, laying the foundation for a high-performance blood test for predicting multiple events in lung cancer.

KEY POINTS: Our study developed an innovative nanomaterial, Zeolite NaY, which addressed the masking effect and improved the depth of the proteome. The performance of NaY-based plasma proteomics as a preclinical diagnostic tool was validated through both internal and external cohort. Furthermore, we explored the different patterns of plasma protein changes during the progression of lung cancer and used the explanations method to elucidate the roles of proteins in the multitask predictive model.

PMID:39783847 | PMC:PMC11714244 | DOI:10.1002/ctm2.70160