AF: Small: Accurate, Biochemically-Relevant, and Robust Scoring Functions for Protein-Ligand Binding Affinity Prediction

蛋白质-配体结合亲和力是许多重要过程的主要决定因素，例如细胞信号传导、基因调控、代谢和免疫，这些过程依赖于蛋白质与某些底物分子的结合。因此，它在药物设计中发挥着核心作用。由于实验性药物发现的高昂成本和延迟，学术界和制药和生物技术公司依赖于使用计算分子对接的虚拟筛选。通常情况下，这涉及到对接的数万至数百万的配体候选人到一个目标蛋白受体？的结合位点，并使用合适的评分函数来评估每个候选物的结合亲和力，以鉴定作为先导物或有希望的蛋白质抑制剂的最佳候选物。由于评分函数（SF）是用来评分，排名，并确定药物线索，保真度与它预测的亲和力的候选配体的蛋白质？的结合位点和它的计算复杂性对虚拟筛选的准确性和通量具有显著的影响。然而，当前最先进的评分函数具有许多缺陷，包括亲和力预测的中等准确度或低通量、不一致的准确度、所提供的准确度-通量权衡的不确定性以及仅依赖于单一类别的评分函数。准确预测大量不同蛋白质的结合亲和力-配体复合物仍然是计算生物分子科学中最具挑战性的问题之一，其在药物发现、化学生物学和结构生物学中具有应用。我们试图通过开发有效的离散优化算法来解决这个问题，这些算法有助于：（1）设计精确的、高通量的单SF和多SF方法，对于给定的蛋白质-配体复合物数据集具有可证明的最优性;（2）通过适当地约束所选择的蛋白质-配体复合物特征的新型生化规则过滤器来确定生化相关的SF;（3）通过新颖的多SF方法的预测鲁棒性，该方法减少了与仅依赖单个SF相关联的准确度的变化;以及（4）通过新的集成动态多SF方法提供的准确度-吞吐量折衷的灵活性。更广泛的影响：该项目将产生一些更广泛的影响：（1）通过促进有效和具有成本效益的药物发现而使公共卫生受益，这反过来又有助于降低药物成本并提高可负担性;（2）对使用评分函数类型方法的其他领域的影响;（3）对学生进行重要应用领域的跨学科培训;（4）传播项目期间开发的研究成果和软件产品;（5）参与和培训代表性不足的群体和K-12外联。

Protein-ligand binding affinity is the principal determinant of many vital processes, such as cellular signaling, gene regulation, metabolism, and immunity, that depend upon proteins binding to some substrate molecule. Consequently, it has a central role in drug design. Due to prohibitive costs and delays associated with experimental drug discovery, academia and pharmaceutical and biotechnology companies rely on virtual screening using computational molecular docking. Typically, this involves docking of tens of thousands to millions of ligand candidates into a target protein receptor?s binding site and using a suitable scoring function to evaluate the binding affinity of each candidate to identify the top candidates as leads or promising protein inhibitors. Since a scoring function (SF) is used to score, rank, and identify drug leads, the fidelity with which it predicts the affinity of a ligand candidate for a protein?s binding site and its computational complexity have a significant bearing on the accuracy and throughput of virtual screening. However, current state-of-the-art scoring functions have a number of deficiencies, including either mediocre accuracy for affinity prediction or low throughput, inconsistent accuracy, inflexibility in accuracy-throughput trade-off provided, and reliance on only a single category of scoring function.INTELLECTUAL MERIT: Accurately predicting the binding affinities of large sets of diverse protein-ligand complexes remains one of the most challenging problems in computational biomolecular science, with applications in drug discovery, chemical biology, and structural biology. We seek to address this problem by developing efficient discrete optimization algorithms that facilitate: (1) the design of accurate, high-throughput single and multi SF methods with provable optimality for a given protein-ligand complex dataset; (2) determination of biochemically-relevant SFs through novel biochemical rule filters that suitably constrain the protein-ligand complex features selected; (3) prediction robustness through a novel multi-SF approach that reduces the variance in accuracy associated with relying on only a single SF; and (4) flexibility in accuracy-throughput tradeoff provided through a new integrated dynamic multi-SF approach. BROADER IMPACTS: This project will have a number of broader impacts: (1) public health benefits by facilitating efficient and cost-effective drug discovery, which in turn helps lower drug costs and improves affordability; (2) impact on other domains where scoring function type approaches are used; (3) interdisciplinary training of students in an important application area; (4) dissemination of research and software artifacts developed during the project; and (5) participation and training of underrepresented groups and K-12 outreach.