High resolution modeling and design of immune recognition

项目摘要： 免疫受体及其识别的精确建模是计算生物学中的一个主要挑战， 与许多疾病和治疗方法直接相关。虽然它们具有共同的异源二聚体免疫球蛋白 折叠，T细胞受体（TCR）和抗体的巨大序列差异导致了惊人的范围 抗原结合模式和特异性。目前的对接方法在很大程度上无法产生 这些复合物的近天然模型在一组排名靠前的预测，和构象的灵活性 TCR和抗体环是预测算法的主要障碍。我的实验室长期以来 对开发和应用算法以更好地建模和设计TCR和抗体感兴趣。我们最近 开发了一种算法和网络服务器，从序列中建模TCR（TCR模型），TCR数据库 结构和序列（TCR 3d），我们已经组装了一个更新的对接基准，这是正在进行的 用于改进TCR对接算法。我们最近还开发了一个更新的 抗体-抗原对接和亲和力基准，其大小是先前基准的两倍多 发布;我们对这些案例进行了对接和亲和力预测评估，为我们提供了丰富的 模型和分数的数据集。在接下来的五年里，我们计划扩大和利用这些数据集 开发先进的基于知识的工具和算法，包括几何深度学习方法， 解决这一领域的主要挑战：可靠的CDR 3环结构建模，准确的预测抗体， 抗原和TCR-肽-MHC对接，以及TCR和抗体靶向设计。这将导致 从序列中建模TCR和抗体相互作用结构，精确控制TCR和抗体亲和力 和特异性，以及设计与目标抗原的新相互作用。我们将公布我们的方法 并将结果作为Web服务器、数据库和代码提供给社区。这项工作将得到加强， 与领先的实验室合作，通过这些实验室，我们将获得新的实验结构， 动态和亲和性数据，将用于开发，应用和验证我们的算法。

Project Summary: Accurate modeling of immune receptors and their recognition is a major challenge in computational biology, of direct relevance to many diseases and therapeutics. While they share common heterodimeric immunoglobulin folds, the immense sequence diversities of T cell receptors (TCRs) and antibodies lead to an astounding range of antigen binding modes and specificities. Current docking approaches are largely incapable of producing near-native models of these complexes in the set of top-ranked predictions, and conformational flexibility of TCR and antibody loops pose a major barrier to predictive algorithms. My laboratory has had a longstanding interest in developing and applying algorithms to better model and design TCRs and antibodies. We recently developed an algorithm and web server to model TCRs from sequence (TCRmodel), a database of TCR structures and sequences (TCR3d), and we have assembled an updated docking benchmark, which is being used to develop improvements to our TCR docking algorithm. We have also recently developed an updated antibody-antigen docking and affinity benchmark, which more than doubles the size of the previous benchmark release; we have performed docking and affinity prediction assessment on these cases, giving us a rich dataset of models and scores. During the next five years, we plan to expand and capitalize on these datasets to develop advanced knowledge-based tools and algorithms, including geometric deep learning methods, to address major challenges in this area: reliable modeling of CDR3 loop structures, accurate predictive antibody- antigen and TCR-peptide-MHC docking, and design of TCR and antibody targeting. This will result in the ability to model TCR and antibody interaction structures from sequence, precise control of TCR and antibody affinity and specificity, and the design of new interactions to target antigens of interest. We will release our methods and results to the community as web servers, databases, and code. This work will be enhanced by collaborations with leading laboratories, through which we will have access to new experimental structural, dynamic, and affinity data which will be used to develop, apply, and validate our algorithms.