Advancing predictive physical modeling through focused development of model systems to drive new modeling innovations

项目摘要/摘要 这项工作旨在推进生物分子设计的定量方法，特别是预测 生物分子相互作用，通过一系列有重点的社区盲目预测挑战。物理方法用于 预测结合自由能，或“自由能法”，将极大地重塑早期药物 发现，并已发现在制药铅优化中的应用。然而，性能是 不可靠，适用范围有限，在制药应用中的失败往往很难 了解并修复。另一方面，这些方法现在通常可以预测各种简单的物理 性质，如溶剂化自由能或相对溶解度，尽管仍有明显的空间 提高了精确度。近年来，竞争和众包已被证明是一种有效的 推动多领域创新。在我们的领域中，盲目预测挑战在驾驶中发挥了关键作用 在物理性能和结合预测方面的创新，特别是以样本系列的形式 挑战。在这里，我们将继续并扩展样本预测挑战，以包括新的物理 性质，更复杂的主客体结合数据，以及在生物分子系统中的应用。 精心选择的系统和新颖的实验数据将带来逐渐增加的挑战 系统之间的复杂性，现在这些系统对那些略微无法达到的系统来说是容易处理的 今天的方法，但仍然比药物设计数据资源(D3R)系列中介绍的方法稍微简单一些 对现有制药数据的挑战。我们将与D3R合作，对我们的数据进行盲目挑战 产生并确保其设计为最大限度地造福于该领域。 在我们最初的提案中，目标4专注于使用在一系列挑战样本中生成的数据，应用 经过验证的基于众包的技术，以推动新方法的开发和对 现有技术的优点和缺点。在这里，我们通过构建软件来扩展这项工作 针对这些挑战的全自动化组件的基础架构，其中工作流组件可以 存放在一个共同的登记处，然后链接在一起，以自动参与SAMPL挑战。这 同时解决了几个关键问题，并将使样本挑战产生的创新具有 对社区的影响要大得多，并更快地传播到各种应用程序。 在样本挑战中使用的软件的用户数量在数千到数万之间，所以这 将对预测建模社区产生深远的影响。

PROJECT SUMMARY/ABSTRACT This work seeks to advance quantitative methods for biomolecular design, especially for predicting biomolecular interactions, via a focused series of community blind prediction challenges. Physical methods for predicting binding free energies, or “free energy methods”, are poised to dramatically reshape early stage drug discovery, and are already finding applications in pharmaceutical lead optimization. However, performance is unreliable, the domain of applicability is limited, and failures in pharmaceutical applications are often hard to understand and fix. On the other hand, these methods can now typically predict a variety of simple physical properties such as solvation free energies or relative solubilities, though there is still clear room for improvement in accuracy. In recent years, competitions and crowdsourcing have proven an effective model for driving innovations in diverse fields. In our field, blind prediction challenges have played a key role in driving innovations in prediction of physical properties and binding, especially in the form of the SAMPL series of challenges. Here, we will continue and extend SAMPL prediction challenges to include new physical properties, more complicated host-guest binding data, and application to biomolecular systems. Carefully selected systems and novel experimental data will provide challenges of gradually increasing complexity spanning between systems which are now tractable to those which are marginally out of reach of today's methods but still slightly simpler than those covered by the Drug Design Data Resource (D3R) series of challenges on existing pharmaceutical data. We will work with D3R to run blind challenges on the data we generate and to ensure it is designed to maximally benefit the field. In our original proposal, Aim 4 focused on using data generated in a SAMPL series of challenges, applying proven crowdsourcing-based techniques to drive the development of new methods and new understanding of the strengths and weaknesses of existing techniques. Here, we extend this work by building out software infrastructure for a fully automated component of these challenges, where workflow components can be deposited in a common registry and then linked together to automate participation in SAMPL challenges. This solves several key problems at once, and will allow innovations resulting from the SAMPL challenges to have much greater impact on the community and much more rapidly disseminate to a wide variety of applications. Users of software employed in the SAMPL challenges number in the thousands to tens of thousands, so this will have far-reaching implications for the predictive modeling community.