針對可支援多種混合模型之微流體生物晶片 並具備多重最佳化目標之樣本製備技術

Abstract

樣本製備程序(sample preparation)乃各種生化反應中極其重要的步驟之一。原始的生物樣本或反應試劑必須在此程序中進行稀釋或混合，以達到反應所需的目標濃度(target concentration)。現今的樣本製備技術大多是架構於僅支援(1:1)混合模型的數位微流體晶片。因為缺乏其他可用的混合模型，此類演算法所得到的結果常不盡理想。現今，支援多種混合模型的通道式微流體生物晶片之兩種針對反應試劑做最小化的樣本製備程序TPG以及VOSPA皆展現了較佳的結果。雖然如此，在進行樣本製備程序時，反應試劑的用量並非永遠皆為使用者的第一考量；有時甚至必須同時考慮所有最佳化目標。因此，本篇論文將提出全球第一篇使用多種混合模型並具備多重最佳化目標之樣本製備技術演算法。此演算法首先會使用動態規劃求出所有可用的混合組合，並利用這些組合建立網路流模型，最後藉由整數線性規劃求得結果。與之前支援多種混合模型的通道式微流體生物晶片的演算法相比較，此演算法無論在樣本試劑及緩衝溶液的用量、廢液多寡、亦或是反應步驟數皆可提供較佳的結果。

Sample preparation is an important process in most biochemical reactions. During this process, reagents are mixed repeatedly to get certain concentration values. Current sample preparation algorithms are mostly designed for digital microfluidic biochips with the (1:1) mixing models. However, the results are generally poor due to the lack of mixing models. Two recently research TPG [27] and VOSPA [26] had shown that the valuable reactant usage can be massively reduced if flow-based microfluidic biochips with various mixing models are adopted. Besides, reactant minimization is not always the first priority during sample preparation. Furthermore, all of them should be considered concurrently. Hence, in this paper, we proposed the first sample preparation algorithm for microfluidic biochips with various mixing models that can perform multi-objective optimization simultaneously. It first enumerate all available mixing combinations via dynamic programming. Then it formulates the problem in a network-flow model with edges are implied by mixing recipes. Finally it is solved through integer linear programming (ILP). Experimental results show that the proposed method can provide better solutions (in terms of reactant, waste, and operation jointly) as compared with the prior art.