近年來許多研究利用基因表現資料來預測基因之間的轉錄調控關係，並加以實驗証實。然而，基因的表現大都經由多轉錄因子共同調控，用傳統的方法並不適合辨識出這種調控關係。我們發展了RECEC演算法，可以較不受多轉錄因子共同調控所造成的混淆因素所影響，來合理的評估基因之間的相關性，藉以預測調控關係。我們用了612片的大腸桿菌生物晶片資料推測基因轉錄調控關係，並且用已知的3,124調控關係評估預測效果，我們演算法預測效果較佳，AUC(ROC)值達到73.74%，而傳統方法達到70.66%。我們並且用膠體位移實驗檢測我們預測的調控關係，証實LexA蛋白質與nac基因上游有鍵結活性，而當限定轉錄因子是LexA蛋白時，這筆預測是我們演算法裡得分最高的一筆，然而在傳統的方法裡卻是第38筆。 Network inference from microarray data has been applied to and eased the task of identifying transcriptional regulatory interactions. However, gene expression is generally controlled by combinatorial interaction of transcription factors (TFs). It’s hard to reconstruct the network properly using the relatedness of gene expression between pairs of genes assessing by traditional methods. Here we developed and applied the Relatedness Estimation under Confounding Effect Control (RECEC) algorithm. Our approach enables a more proper estimation of the relatedness with less confounding effect resulted from combinatorial regulation of TFs. We inferred the network from 612 Escherichia coli microarray data and evaluated the inference performance using known 3,124 transcriptional regulatory interactions. Our algorithm demonstrates a better AUC(ROC) 73.74% compared to traditional approach 70.66%. We also conducted EMSA experiments to indentify putative transcriptional regulatory interactions inferred by our algorithm. We found TF LexA binds to the upstream region of nac gene. The relatedness of this interaction is ranked number 1 in our algorithm compared to number 38 in traditional methods when TF is restricted to LexA. Our approach offers the potential to identified novel transcriptional regulatory interactions which are involved in combinatory regulation of transcription.
