Scholarlink

Background <p>Resistance remains a major barrier to targeted cancer therapies. Axitinib, a VEGF receptor inhibitor with anti-angiogenic activity, is effective in several cancers but shows heterogeneous patient responses, reflecting context-specific molecular adaptations. A comprehensive multi-omics approach is needed to define these mechanisms and uncover compensatory survival pathways limiting Axitinib efficacy.</p> Methodology <p>We conducted a high-throughput analysis of ~ 1000 pan-cancer cell lines treated with 44 FDA-approved targeted drugs. Basal transcriptomic (~ 36,000 transcripts) and proteomic (~ 9000 proteins) profiles were integrated to predict cell-line-specific drug response using a multi-classifier machine learning framework. Multiple models, including ensemble, linear, and kernel-based classifiers, were trained per drug and evaluated via fivefold cross-validation. Axitinib, the best-predictive drug, was further analyzed using explainable AI (LIME) to identify resistance-driving features for each cell line. Resistant cell lines were clustered using agglomerative hierarchical clustering based on LIME-identified features and highly correlated partners. Optimal clusters were determined via silhouette scoring. Enrichment analysis, pathway annotation, and literature mining were used to uncover cluster-specific resistance mechanisms.</p> Results <p>Axitinib achieved the highest predictive accuracy across all 44 drugs. The machine learning pipeline reliably classified cell lines as resistant or sensitive from basal transcriptomic and proteomic data. LIME identified key resistance-driving features at the individual cell line level. Clustering based on these features revealed two resistance subtypes shaped by tissue origin and survival constraints. In blood-derived cancers, resistance involves purine metabolism rewiring and alternative growth factor signaling to sustain proliferation. In solid tumors, resistance reflected adaptation to hypoxia, including ECM remodeling, mechanosensing, EMT, immune evasion, and senescence-induced paracrine signaling.</p> Conclusion <p>Axitinib resistance emerges through tissue- and context-specific adaptations. Multi-omics profiling with explainable machine learning reveals distinct survival strategies, underscoring the need for precision re-sensitization approaches tailored to tumor context.</p> Graphical Abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deciphering context-specific Axitinib escape pathways via multi-omics and explainable machine learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月12日

查看全文
  • Samriddhi Gupta,
  • Khyati Patni,
  • Simarpreet Kaur,
  • Jaspreet Kaur Dhanjal

摘要

Background

Resistance remains a major barrier to targeted cancer therapies. Axitinib, a VEGF receptor inhibitor with anti-angiogenic activity, is effective in several cancers but shows heterogeneous patient responses, reflecting context-specific molecular adaptations. A comprehensive multi-omics approach is needed to define these mechanisms and uncover compensatory survival pathways limiting Axitinib efficacy.

Methodology

We conducted a high-throughput analysis of ~ 1000 pan-cancer cell lines treated with 44 FDA-approved targeted drugs. Basal transcriptomic (~ 36,000 transcripts) and proteomic (~ 9000 proteins) profiles were integrated to predict cell-line-specific drug response using a multi-classifier machine learning framework. Multiple models, including ensemble, linear, and kernel-based classifiers, were trained per drug and evaluated via fivefold cross-validation. Axitinib, the best-predictive drug, was further analyzed using explainable AI (LIME) to identify resistance-driving features for each cell line. Resistant cell lines were clustered using agglomerative hierarchical clustering based on LIME-identified features and highly correlated partners. Optimal clusters were determined via silhouette scoring. Enrichment analysis, pathway annotation, and literature mining were used to uncover cluster-specific resistance mechanisms.

Results

Axitinib achieved the highest predictive accuracy across all 44 drugs. The machine learning pipeline reliably classified cell lines as resistant or sensitive from basal transcriptomic and proteomic data. LIME identified key resistance-driving features at the individual cell line level. Clustering based on these features revealed two resistance subtypes shaped by tissue origin and survival constraints. In blood-derived cancers, resistance involves purine metabolism rewiring and alternative growth factor signaling to sustain proliferation. In solid tumors, resistance reflected adaptation to hypoxia, including ECM remodeling, mechanosensing, EMT, immune evasion, and senescence-induced paracrine signaling.

Conclusion

Axitinib resistance emerges through tissue- and context-specific adaptations. Multi-omics profiling with explainable machine learning reveals distinct survival strategies, underscoring the need for precision re-sensitization approaches tailored to tumor context.

Graphical Abstract