FET: CCF: Small: Computational Drug Prediction through Joint Learning

传统的基于靶标的药物开发方法耗时长，成本高，失败率高。随着有关药物、疾病和药物靶点的数据越来越多，能够整合各种异构数据源集的计算方法在加速药物开发过程方面具有巨大潜力。该项目通过开发先进的人工智能和机器学习算法，并将其应用于来自不同领域的综合真实数据，解决了计算药物预测中的基本问题。该项目还将通过面向广大学生群体（包括代表性不足的群体）的教育和推广活动，支持扩大对计算机的参与。更具体地说，该项目将通过强大的计算方法研究药物、疾病和靶点之间的关系，包括张量和张量分解、多视图学习和深度学习，以实现药物的合理再利用。该项目将侧重于开发高效和有效的学习算法，对算法收敛性和复杂性进行严格的理论分析，并使用从各种数据库获得的真实数据进行综合评估。该方法将张量分解与多视图学习和深度学习无缝结合。通过在多视图学习框架中利用辅助信息，可以有效地解决稀疏性问题，使学习到的隐藏结构更有意义和可解释性。神经张量分解方法将允许探索潜在特征之间更复杂的非线性关系。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Traditional target-based drug-development approaches are lengthy and costly, with high failure rates. With increasingly available data regarding drugs, diseases, and drug targets, computational approaches that can integrate diverse sets of heterogeneous data sources have great potential to speed up the drug-development process. This project addresses fundamental questions in computational drug prediction by developing advanced AI and machine-learning algorithms and applying them to comprehensive real data from different domains. The project will also support broadening participation in computing via educational and outreach activities geared towards a wide group of students, including underrepresented groups.More specifically, the project will study the relationships among drugs, diseases, and targets via powerful computational methods including tensors and tensor decomposition, multi-view learning, and deep learning for rational drug repurposing. The project will focus on the development of efficient and effective learning algorithms, rigorous theoretical analysis of algorithm convergence and complexity, and comprehensive evaluations using real data obtained from various databases. The approach will seamlessly integrate tensor decomposition with multi-view learning and deep learning. By utilizing auxiliary information in the framework of multi-view learning, it can effectively address the sparsity issue, and the learned hidden structures should be more meaningful and interpretable. The neural tensor-decomposition approach will allow exploration of more complicated nonlinear relationships among latent features.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.