腐蝕抑制劑可用於保護金屬表面，防止在特定環境中腐蝕反應的發生，廣泛應用於石油與天然氣產業等領域。對於加速高腐蝕效率的鐵（Fe）腐蝕抑制劑的開發，是目前該領域中的一項重大挑戰。 在本研究中，我們提出一個以機器學習為基礎的預測框架，用於預測有機腐蝕抑制劑在鹽水環境中鐵表面上的抑制效率（IE）。此框架結合來自最穩定互變異構體結構之量子化學與結構描述符進行建模。我們採用了三種機器學習演算法，包括人工神經網路（ANN）、Extreme Gradient Boosting（XGBoost）與隨機森林（RF）建立預測模型。在這三種模型中，ANN 展現最佳效能，在 10 次交叉驗證中達到 7.00% 的平均絕對誤差（MAE）。在包含近期發表的數據與我們的實驗測量結果的獨立測試集上，模型達成 9.32% 的 MAE。 該模型亦應用對原來針對銅、鋁、鎂及於鹽酸或硫酸溶液環境中所開發的 588 個抑制劑進行虛擬篩選，成功識別出 265 個分子在鹽水溶液中對鐵預測 IE 高於 90% 的潛力候選者，展現其作為抑制劑再定位工具的潛力。透過 SHAP 分析，我們揭示了影響抑制效能的關鍵電子與結構特徵，而片段統計分析則指出雙環結構（如苯並咪唑與 8-羥基喹啉）為高 IE 的主要貢獻片段。此整合式方法實現快速且具成本效益的腐蝕抑制劑開發，並為未來分子工程設計提供寶貴依據。 ;Corrosion inhibitors are essential for safeguarding metal surfaces by suppressing corrosion processes under various environmental conditions. Their use is especially prevalent in sectors like oil and gas. However, developing highly effective corrosion inhibitors for iron (Fe) remains a significant challenge in the field. In this work, we propose a machine learning (ML)-based predictive framework aimed at estimating the inhibition efficiency of corrosion inhibitors on iron surfaces in a NaCl medium. The model integrates quantum chemical and structural descriptors derived from the most stable tautomeric forms of the molecules. To construct the predictive models, we employed three ML algorithms: Artificial Neural Networks, Extreme Gradient Boosting, and Random Forest. Among them, the ANN model demonstrated the highest accuracy, achieving a mean absolute error of 7.00% through 10-fold cross-validation. When evaluated on an independent test set comprising newly reported data and our own experimental results, the model achieved an MAE of 9.32%. We further applied the trained model to perform virtual screening on a set of 588 known inhibitors originally designed for aluminum, magnesium, and iron under acidic conditions (HCl or H₂SO₄). The model successfully identified 265 compounds with predicted IE values over 90% in a NaCl environment for iron, indicating strong potential for repurposing these inhibitors. Additionally, SHAP analysis was used to uncover the key electronic and structural factors influencing inhibitory performance, while fragment enrichment analysis highlighted bicyclic structures—such as benzimidazole and 8-hydroxyquinoline—as critical motifs associated with high efficiency. This combined approach offers a fast and cost-effective strategy for discovering novel corrosion inhibitors and serves as a valuable tool for guiding future molecular design efforts.
