二次二進位最佳化於分子篩選之應用

Abstract

在眾多可能性中篩選分子是一項具挑戰性的任務。二次無約束二進制優化(QUBO) 求解器的出現，為解決這一類的問題提供了不同的替代方法。我們開發出了一種將 QUBO 求解器與密度泛函理論計算相結合的分子篩選流程。在這項概念性驗證的工作中，我們將問題聚焦於篩選酚類抑制劑，並將問題映射到 QUBO 形式。其中，酚類 O-H 鍵的鍵解離能是酚類抑制劑有效的關鍵指標。為此，我們的方法使用基團貢獻法將酚類 O-H 鍵的鍵解離能近似成 QUBO 形式。我們的結果顯示，QUBO 模型所近似出的鍵解離能數值與密度泛函理論計算所得出的數值之間有很強的關聯性，相關係數達到 0.82, 而斯皮爾曼相關係數達到了 0.86。如此的高相關性確保 QUBO 求解器能夠有效識別潛在候選分子。基於此 QUBO 模型，我們使用 D-Wave 量子退火器和富士通退火機器來篩選出候選分子。此外，我們還通過密度泛函理論的計算驗證 QUBO 求解器所篩選出的候選分子的有效性。這項工作提供了能結合基團貢獻法和 QUBO 求解器的前瞻應用方向。

Screening molecules from numerous possibilities is a challenging task. The advent of quadratic unconstrained binary optimization (QUBO) solvers provides an alternative to address this issue. We have developed a process for screening molecules that integrates QUBO solvers with density functional theory (DFT) calculations. As a proof-of-concept, we map the problem of screening phenolic inhibitors onto the QUBO form. In our approach, we approximate the bond dissociation energy (BDE) of the phenolic O-H bond–a key indicator of effective polymeric inhibitors–within the QUBO model, relying on adapting the Group Contribution Method (GCM). Our results demonstrate a strong correlation between the BDE values predicted from the QUBO model and DFT calculations, achieving a correlation coefficient of 0.82 and Spearman’s coefficient of 0.86. This high correlation ensures the QUBO solver to identify potential candidates efficiently. We benchmarked the performance of the QUBO solvers–D-Wave quantum annealer and Fujitsu annealer–on solving this phenolic QUBO problem. We also provide the validation results of the screening candidates from the QUBO solvers through DFT calculations. Our work provides a promising direction for incorporating the GCM into QUBO solvers to tackle molecule screening problems.