Adakite-like rocks, as an important recorder of magmatic evolution in mantle and crust, are closely associated with major porphyry Cu deposits. However, the underlying mechanism connecting these associations remains insufficiently elucidated. This study compiles ca. 7000 whole-rock geochemical data from Phanerozoic adakite-like rock samples and from ore-forming porphyries in porphyry Cu deposits. Machine learning investigations are performed to characterize the geochemical variations of adakite-like rocks from various orogenic or cratonic systems and investigate any geochemical similarities with rocks associated with porphyry ore systems. Principal component analysis (PCA), along with t-distributed Stochastic Neighbor Embedding (t-SNE), highlights elemental distinctions between adakite-like rocks formed in mature and juvenile crust and the similarities of rocks from porphyry CuMo deposits with the former and rocks from porphyry CuAu deposits with the latter. Extreme Gradient Boosting (XGBoost) and Support Vector Machine (SVM) analyses are applied to the original dataset and the PCA data derived from the dataset to discriminate adakite-like rocks from various geodynamic settings from each other and from rocks associated with porphyry ore deposits. Our models show that XGBoost and SVM can identify the tectonic regions of adakite-like rocks and rocks from porphyry copper systems with high efficiency and confidence e.g., accuracy = 0.88–0.94 and area under curve (AUC) = 0.69–0.86. The application of SHapley Additive exPlanations (SHAP) values is employed to elucidate the parameters that discriminate the different environments, providing insight into the underlying petrogenetic processes, including diverse levels of formation depth and extent of mantle interaction. Additionally, when examining ore-forming porphyries and adakite-like rocks, the incorporation of heatmaps into taxonomic classification (HTC) provides significant distinguishing factors with substantial geological implications. These findings suggest potential geological processes (e.g., remelting of Cu-sulfide-bearing cumulates or flotation of Cu-sulfide phases attached to vapor bubbles) that may enrich the source magma of ore-forming porphyries beyond the conventional geological processes responsible for adakite-like rock formation, which could serve as an exploration indicator.

中文翻译：

---

揭示全球亚达基石状岩石与斑岩铜矿床之间的联系


Adakite 类岩石作为地幔和地壳岩浆演化的重要记录者，与主要斑岩 Cu 矿床密切相关。然而，连接这些关联的潜在机制仍未充分阐明。本研究从显生代亚金石状岩石样品和斑岩铜矿床中的成矿斑岩中汇编了大约 7000 个全岩地球化学数据。进行机器学习调查是为了表征来自各种造山或克拉通系统的亚达基特状岩石的地球化学变化，并研究与斑岩矿系统相关岩石的任何地球化学相似性。主成分分析 （PCA） 以及 t 分布随机邻域嵌入 （t-SNE） 突出了在成熟和幼年地壳中形成的亚达岩状岩石之间的元素差异，以及斑岩 CuMo 矿床岩石与前者的相似性，斑岩 CuAu 矿床的岩石与后者的相似性。将极端梯度提升 （XGBoost） 和支持向量机 （SVM） 分析应用于原始数据集和从数据集派生的 PCA 数据，以区分来自各种地球动力学设置的亚达基特状岩石彼此以及与斑岩矿床相关的岩石。我们的模型表明，XGBoost 和 SVM 可以高效、可靠地识别亚达基石状岩石和斑岩铜系统中的岩石的构造区域，例如精度 = 0.88–0.94 和曲线下面积 （AUC） = 0.69–0.86。SHapley 加法解释 （SHAP） 值的应用用于阐明区分不同环境的参数，提供对潜在岩石成因过程的见解，包括不同层次的形成深度和地幔相互作用的程度。 此外，在检查成矿斑岩和亚达基特状岩石时，将热图纳入分类分类 （HTC） 提供了具有重大地质意义的重要区分因素。这些发现表明潜在的地质过程（例如，含硫化铜累积物的再熔化或附着在蒸汽气泡上的硫化铜相的浮选），这些过程可能会富集成矿斑岩的源岩浆，超出导致亚德晶状岩石形成的常规地质过程，这可以作为勘探指标。