Increasing the Coverage and Accuracy of CATH for Comparative Genomics and Variant Interpretation

进化产生了蛋白质结构域家族，其中亲属通过同一基因组中的物种形成事件或复制事件联系在一起。广泛的结构域复制和洗牌使多结构域蛋白根据结构域组成具有不同的功能。CATH分类以结构域为主要进化单位，对结构和序列模式明显相似的亲缘生物进行分类。目前有5500个CATH超家族，包含9300万个域名。之前的资金使我们能够大大增加CATH的领域数量。我们希望继续增加这些数据——随着新技术使解决结构和捕获序列数据变得更容易，预计会有更大的扩展。我们将通过与其他分类专家（SCOP的Alexey Murzin）合作，在欧洲生物信息学研究所建立一个新领域的共享领域识别平台，提高我们领域数据的准确性，并由CATH/SCOP专家共同验证困难的任务。这些数据将是公开的，对其他资源（如SCOPe、ECOD）有价值。CATH已经建立了22年，以提供准确的生物分析结构注释而闻名。最近，它通过提供功能预测显着增加了其对生物界的价值。虽然超家族的结构核心是高度保守的，但远离核心的变异会导致功能的变化。CATH通过将可能具有高度相似功能和结构的进化亲属分组到功能家族（FunFams）来解决这个问题。因此，FunFams可以在亲属之间准确地继承有关结构和功能的信息。这一点很重要，因为只有不到10%的结构域被实验表征。我们在计算机上验证了FunFams可以准确地模拟未表征亲属的结构，并且FunFams在亲属之间继承功能信息的能力已被国际竞争- CAFA验证。我们将使FunFams更加全面，并提高FunFams对酶的准确性。扩展我们的FunFam库将使我们能够在基因组序列中预测更准确的多域注释。这将有助于生物学家比较占据不同环境生态位的生物体的基因组，因为识别不同的结构域组合可以暗示生物体功能库的变化以及在其环境中利用化合物的不同能力。因为《FunFams》中的亲属在结构上是如此保守，我们可以将它们排列和叠加以提取保守结构核心的特征，并使用这些信息来构建“3D核心模板”。由于强大的新结构生物学技术（如冷冻电镜）可以使用这些核心文库来模拟电子弥散数据中未表征的蛋白质的结构，这些模板将有助于解决更多亲缘关系的结构问题。在CATH的另一个令人兴奋的发展中，我们将利用结构数据和来自200倍大的序列数据的额外力量来寻找蛋白质中的残基位点，这些残基位点在整个进化过程中因其功能重要性而保守。我们将描述这些地点的特点。我们已经从序列数据的保守模式中很好地预测了功能位点，但包括结构数据可以帮助区分位点的类型（例如结合化合物或另一种蛋白质的位点）并识别参与功能机制的其他残基。这些数据对于蛋白质设计和理解为什么这些位点附近的突变会影响蛋白质并导致疾病是有价值的。我们将通过网页和其他网络机制传播我们的数据，并为新功能制作电子视频和培训材料。我们还将建立更有效的机制来扫描我们的网站，让生物学家在他们自己的电脑上安装我们的工具来分析基因组数据。

Evolution has given rise to families of protein domains where relatives are linked through speciation events or duplication events in the same genome. Extensive domain duplication and shuffling gives multi-domain proteins with varying functions depending on the domain composition.The CATH classification takes the domain as the primary evolutionary unit and classifies relatives having significantly similar structures and sequence patterns. Currently there are 5500 CATH superfamilies containing 93 million domains. Previous funding allowed us to hugely increase the number of domains in CATH. We want to keep increasing this data - even bigger expansions are expected as new technologies make it easier to solve structures and capture sequence data. We will improve the accuracy of our domain data by working with other classification experts (Alexey Murzin of SCOP) to establish a shared domain recognition platform for new domains at the European Bioinformatics Institute, with difficult assignments jointly validated by CATH/SCOP experts. This data will be public and valuable for other resources (eg SCOPe, ECOD).CATH has been established for 22 years and is renowned for providing accurate structural annotations for biological analyses. More recently it significantly increased its value to the biology community by providing functional predictions. Although the structural core of the superfamily is highly conserved, variations away from the core cause changes in function. CATH addresses this by grouping evolutionary relatives likely to have highly similar functions and structures into functional families (FunFams). Thus FunFams can accurately inherit information about structures and functions, between relatives. This is important as <10% of domains have been experimentally characterised. We verified in-silico that FunFams can accurately model structures of uncharacterised relatives and the ability of FunFams to inherit functional information between relatives has been validated by an international competition - CAFA. We will make the FunFams much more comprehensive and increase the accuracy of FunFams for enzymes.Extending our FunFam library will allow us to predict more accurate multi-domain annotations in genome sequences. This will help biologists comparing the genomes of organisms occupying different environmental niches, as identification of diverse domain combinations can hint at changes in the functional repertoires of the organisms and different abilities to exploit compounds in their environments.Because relatives in FunFams are so structurally conserved we can align and superpose them to extract the characteristics of this conserved structural core and use this information to build a '3D core-template'. These templates will help solve the structures of many more relatives since powerful new structural biology techniques (eg cryo-EM) can use core libraries like these to model the structures of uncharacterised proteins from electron dispersion data.In another exciting development for CATH we will harness the structural data and the additional power that comes from 200-fold greater sequence data to find residue sites in the protein, conserved throughout evolution for their functional importance. We will characterise these sites. We already predict functional sites well from conservation patterns in sequence data, but including structural data can help distinguish the type of site (eg site binding a compound or another protein) and identify additional residues involved in the functional mechanism. This data is valuable for protein design and understanding why mutations near these sites affect the protein and cause disease.We will disseminate our data via webpages and other web mechanisms and develop e-videos and training material for the new features. We'll also build more efficient mechanisms for scanning our website and for biologists to install our tools on their own computers to analyse genome data.