VISUALIZING THOUSAND-DIMENSIONAL CHEMICAL DIVERSITY

这项工作的目标是提供交互式计算机可视化，研究科学家可以使用它来解释高通量筛选数据，并做出组合化学选择。最简单的药物发现原则是，性质相似的化合物通常在生物活性上也相似。相似性通常涉及高维空间中的度量，例如分子指纹或形状描述符。相似性在药物发现研究中的应用可能适用于来自潜在可合成化合物虚拟图书馆的数百万种化合物。通过与空间的二维地图进行简单的图形交互，来检查多样性空间中大量化合物之间的关系，允许经验丰富的科学家的直觉发挥作用。千维多样性空间的可视化算法依赖于水平，即距离矩阵不需要求解的距离，以及有效的子采样方法。当组合在遗传算法中时，这些概念还使得能够选择最佳描述符来聚类化合物以供预测使用。最佳的描述符不仅有助于可视化多样性空间的重要特征，而且有助于在活性物质的早期类比过程中决定下一步要制造和测试的化合物。拟议的商业应用：执行多样性选择的软件需要这些可视化工具。为随机筛选或跟踪热门歌曲而提供的复合库，当它们的设计能够得到说明时，就更有价值了。这些工具适用于差异基因表达数据分析等新领域。分析HTS数据的新方法具有改善早期药物发现研究过程的商业潜力。

The goal of this work is to provide interactive computer visualizations which research scientists can use to interpret high throughput screening data and to make combinatorial chemistry choices. The simplest drug discovery principle is that compounds similar in enough properties are usually similar in biological activity. Similarity often involves measures in high-dimensional spaces, such as molecular fingerprints or shape descriptors. Uses of similarity in drug discovery research may apply to millions of compounds from virtual libraries of potentially synthesizable compounds. To examine relationships among vast numbers of compounds in diversity space, by simple graphical interactions with two dimensional maps of the space, allows the intuition of experienced scientists to come into play. The algorithms for visualization of thousand dimensional diversity spaces rely on horizons, which are distances beyond which the distance matrix need not be resolved, and on efficient subsampling methods. These concepts also enable selection of optimal descriptors to cluster compounds for predictive use, when combined in genetic algorithms. Optimal descriptors help not only in visualizing important features of diversity space, but in deciding which compounds to make and test next during early analoging of active substances. PROPOSED COMMERCIAL APPLICATION: Software that performs diversity selections needs these visualization tools. Compound libraries offered for random screening or following up on hits are more valuable when their designs can be illustrated. The tools apply to new areas such as differential gene expression data analysis. New methods for analyzing HTS data have commercial potential of improving the process of early drug discovery research.