2026-10-06T14:38:44Z https://api.figshare.com/v2/oai

oai:figshare.com:article/33653113 2026-09-12T07:01:57Z category_24187 category_24196 item_type_3 month_year_09_2026

Interpretable AI Fusion Prioritizes Nomilin-Responsive Targets and PI3K/Akt/Cyclin D1 Signaling in Triple-Negative Breast Cancer Shaojun Li (24378059) Biological network analysis Proteomics and metabolomics artificial intelligence triple-negative breast cancer nomilin target prioritization graph embedding XGBoost positive-unlabeled learning network pharmacology proteome microarray PI3K/Akt pathway Triple-negative breast cancer (TNBC) remains difficult to treat because of its molecular heterogeneity, aggressive progression, and limited actionable targets. Nomilin, a citrus-derived limonoid, has shown potential antitumor activity, but its target network and mechanism in TNBC remain incompletely characterized. This study developed an interpretable artificial intelligence (AI)-assisted multi-source evidence fusion framework to prioritize nomilin-responsive targets and validate their biological relevance. In Stage I, transcriptomic dysregulation, weighted gene co-expression modules, disease-target evidence, compound-target evidence, protein-protein interaction topology, graph embedding, pathway context, and molecular docking were integrated into a positive-unlabeled (PU) ensemble target-ranking framework with an independent held-out reference set. In Stage II, molecular dynamics simulation, MM/GBSA binding free energy, proteome microarray, cellular thermal shift assay, microscale thermophoresis (MST), single-cell mapping, immune deconvolution, cellular assays, Akt rescue, and xenograft validation were used as independent post-ranking evidence. Among 221 candidate nomilin-TNBC targets, PARP1, EGFR, HSP90AA1, BCL2, CASP3, CCND1, KRAS, and TNF were prioritized as high-priority targets. SHAP analysis showed that graph-derived features, transcriptomic relevance, docking plausibility, and PI3K/Akt pathway context contributed substantially to ranking. Independent held-out evaluation yielded,, and. An external curcumin-colorectal-cancer benchmark showed a smaller but consistent advantage of the PU-XGBoost framework over GraphSAGE and random forest. Post-ranking validation supported protein-level binding signals and selected cellular target engagement. Orthogonal MST measurements supported measurable nomilin binding to several prioritized proteins, whereas TNF showed weak, non-saturating binding within the tested concentration range. Functionally, nomilin inhibited TNBC cell proliferation and migration, induced mitochondria-dependent apoptosis, and suppressed xenograft growth. SC79-mediated Akt activation partially reversed nomilin-induced inhibition of cell viability and Cyclin D1 expression, supporting the involvement of PI3K/Akt/Cyclin D1 signaling. These findings support further investigation of nomilin as a multi-target natural product candidate in TNBC models. 2026-09-12T07:01:57Z Dataset Dataset 10.6084/m9.figshare.33653113.v1 https://figshare.com/articles/dataset/_b_Interpretable_AI_Fusion_Prioritizes_Nomilin-Responsive_Targets_and_PI3K_Akt_b_b_Cyclin_D1_b_b_Signaling_in_Triple-Negative_Breast_Cancer_b_/33653113 CC BY 4.0