Chapter 1 covers the exciting developments in ab initio modelling. This technique is increasingly enabling accurate fold predictions, and even modelling of atomic details in some cases, for small proteins lacking recognisable similarity to known structures. Given recent results, it is startling now to think that, as recently as 1997, Arthur Lesk said, after assessing the CASP 2 ab initio submissions, that “I found the results … disappointing, even sobering, and many people share this reaction. Except for one target, all predictions were limited to no more than fragmentary success” (Lesk 1997). Chapters 2 and 3 focus on structure inference and modelling based on known structures. Comparative modelling, dealt with in Chapter 3, is a mature and important technology capable of consistently providing reliable models in many cases. Just as importantly, it is largely predictable which parts of comparative models are likely to be more or less reliable. Chapter 2 is concerned with recognising folds for protein sequences, often highly valuable information in itself (Chapter 6), where simple sequence comparisons are uninformative. However, fold recognition, just as importantly, extends the boundaries for comparative modelling, in effect increasing the number of models that can be built per experimental structure. For membrane proteins, covered in Chapter 4, structural bioinformatics approaches have been fundamentally limited by the small number of available experimental protein structures. For this reason, the chapter also covers in detail approaches to predicting membrane topology for different classes of protein, effectively a lower-resolution form of structure prediction. Chapter 5 deals with the fascinating class of proteins which show intrinsic structural disorder in isolation, at least in part, but which undergo ordering upon interaction with other molecules. These proteins, research on which has only blossomed in the last decade, have their own idiosyncratic rules of structure-function and so Chapter 5 also covers how function may be predicted for them.

The second part of the book, From Structures to Functions, opens with Chapter 6, a discussion of how function varies and evolves in the context of a superfamily or protein fold. Importantly for structure-based prediction efforts, some folds, when observed or inferred, are reliable indicators of function; others, the super-folds, support diverse functions. Interactions of proteins with ligands necessarily occur at the protein surface so it is no surprise that there are many aspects of its geometry and properties that can usefully be exploited for function prediction. These methods are covered in Chapter 7. Chapter 8 covers the local structural patterns that may be intimately associated with binding or catalysis. Such patterns arise through conservation or convergent evolution of effective catalytic sites, as well as from constraints on binding imposed by the physicochemical characteristics of a given small molecule. Chapter 8 also covers recent success in the use of small molecule docking to predict enzyme specificity. An often overlooked facet of protein function is its relationship to protein dynamics. Protein structures are not static and movements, large or small, are often key to function. Molecular Dynamics and related methods for conformational sampling and analysis are covered in Chapter 9, which also provides examples to show how consideration of dynamics enhances our understanding of protein function. With an ever-increasing number and range of methods for prediction function from structure, it is sensible to provide for application of multiple methods within integrated servers. This is more convenient for the user and also allows for determination of consensus predictions. Chapter 10 describes the resources and operation of ProFunc and ProKnow which work in this way. Chapter 11 discusses published work in which structure-based methods were applied to predict functions for Structural Genomics outputs. This results in a valuable picture of which methods typically prove most informative. The chapter concludes with a discussion of recent moves towards community annotation as a way to tackle the bottleneck in annotation of Structural Genomics results. Chapter 12 covers applications of structure-based methods to model structures, derived by both comparative and ab initio techniques. As well as a broad range of examples, published work assessing the accuracy of model structures in function-relevant aspects and the applicability of various methods to model structures is discussed.