在分析高維基因表現資料時,實驗分批進行常會產生系統性的批次效應(batch effects)。目前廣泛使用的 ComBat 經驗貝氏方法 (Johnson2007) 於第一階段依賴普通最小平方法(OLS)估計生物效應;然而,在基因數遠大於樣本數(G>>n)的高維情境下,逐基因 OLS 估計較易受背景雜訊影響,產生高變異與極端估計值,進而降低標準化殘差矩陣的穩定性與後續批次參數估計之表現。為此,本研究提出{\bf 高維基因表現資料之 ComBat-TPBN 批次效應校正模型},將三參數貝塔常態先驗(Three-Parameter Beta Normal, TPBN)引入 ComBat 第一階段。模型採用全局矩陣視角,先透過組內中心化預先控制批次造成的加法偏移,建構等價資料矩陣;再對其生物效應施加全局--局部收縮先驗,在保留真正訊號的同時,將多數背景基因的估計值收縮至接近零。處理後之標準化殘差可直接銜接原始 ComBat 的經驗貝氏收縮與資料還原機制,藉此提升高維環境下之結構穩定性。本研究透過蒙地卡羅數值模擬與 TCGA 乳癌(BRCA)資料評估模型表現。模擬結果顯示,在高維設定下,TPBN 能有效將真實值為零之背景基因係數收縮至接近零,於此類基因上達到較高判斷正確率,並同時降低整體估計標準誤與最終資料還原的均方誤差。在實證分析中,ComBat-TPBN 一方面保留具有文獻支持之腫瘤驅動基因與表現標記的生物效應估計,另一方面將缺乏實證支持或高度異質之背景基因係數明顯收縮,並有效緩解批次分群現象。整體而言,本模型在保留原有資料還原架構的前提下,提供了一種適用於高維基因表現資料的穩健批次效應校正途徑。;High-dimensional gene expression data are often generated in multiple experimental batches, leading to systematic batch effects. The widely used ComBat empirical Bayes method(Johnson2007) relies on gene-wise ordinary least squares (OLS) in its first stage; in high-dimensional settings (G>>n), these OLS estimates are sensitive to background noise and can produce high-variance, extreme coefficients that destabilize the standardized residuals used for batch adjustment. To address this, we propose \textbf{a ComBat-TPBN batch-effect correction model} for high-dimensional gene expression data. From a global matrix perspective, the model applies within-batch mean-centering to remove additive batch shifts, then imposes a Three-Parameter Beta Normal (TPBN) global–local shrinkage prior on the biological effects so that most background gene coefficients are shrunk toward zero while true signals are preserved, and the resulting standardized residuals are processed by the original ComBat empirical Bayes and data-reconstruction steps. Monte Carlo simulations and an application to TCGA breast cancer (BRCA) data demonstrate the benefits of this approach: in simulations, TPBN shrinks truly null background genes toward zero and reduces both estimation standard errors and reconstruction mean squared error; in the real-data analysis, ComBat-TPBN preserves literature-supported tumor driver effects while shrinking unsupported background genes.