Biological Sequence Analysis : Probabilistic Models of Proteins and Nucleic Acids

Rating:
Summary: Don't let the title mislead you.
Review: Don't let the title fool you. This book is a great if you'd like to understand the algorithms used in any type of sequence analysis, for example speech recognition, speech synthesis, and natural language understanding.

I used this book for a bioinformatics class. The instructor's notes were basically a rehash of the textbook. This didn't bother me as there really is no way to improve on what's already in the text. Explanations of the different ways to use HMMs made it easy to write the genefinder we did for our final programming project.

I've also written natural language processing software (for text and speech) and I've found this book to be a great reference for probabilistic language modeling algorithms. The material is similar to that found in Jurafsky and Martin, or Manning and Schutz, but the presentation in DEKM provides more insight into how the algorithms work. This should come as no surprise, as the human genome project is perhaps the most successful artificial intelligence project ever undertaken and the authors were instrumental in creating the software used by the HGP.

The book by Gusfield is also great for sequence analysis, but there the emphasis is on deterministic modeling, which has it's place if one can't make a probabilistic sequence model.

Mining databases of text, image, and sound sequences is becoming more important as more data is available on the web. Books like DEKM are valuable algorithm resources for extracting knowledge all sorts of sequence data.

Rating:
Summary: Surprisingly deep and clear book, even viewed from outside
Review: I am a physicist and had some interest in what these bio informatics actually do. I must say I am impressed both in the rigor and sharpness of the probabilistic reasoning. This book relies heavily on probability theory (especially hidden Markov models) and is clear enough to be read without a sharp pencil. Don't get me wrong it is not simple enough to be good late night bedtime entertainment. The biological and chemical background is also easy to grasp.
The authors are obviously very active in the field they describe. Their self citations seem absolutely reasonable.

.

Rating:
Summary: Brief and clear
Review: I keep coming back to this book for its readable, applicable summaries of basic algorithms.

One chapter covers the basics of dynamic programming for string matching: a staple of bioinformatics computing. The authors come back to it a number of times as they introduce new variations on the string-matching theme. They give about the clearest description of the Needleman-Wunsch and basic variants (including Smith-Waterman) of any book I know.

The bulk of the book is devoted to Hidden Markov Models (HMMs), as one might have guessed in a book with Eddy as co-author. It covers the basics of model construction, motif finding, and various uses for decoding. Again, it covers all the basics so clearly you'll want to start coding as soon as you read it.

The later sections of the book cover phylogeny and tree building, along with the relationships to multiple alignment. Good, solid, clear writing prepares the reader for texts that may be more specialized, but possibly less transparent.

The next-to-last chapter, on RNA folding, is weaker than the ones before, in my opinion. It ties to the other chapters reasonably well in terms of algorithms, but I don't think it does justice to the thermodynamic models of RNA folding. If there is any weakness in this chapter, though, it does not detract from the strengths elsewhere.

The final chapter, the "background on probability", is the one that I think needs the most support. If you don't already understand its topics, I doubt that this will help very much. (If you do understand them, you won 't need the help.) There's nothing inherently tricky about probability, but individual distributions carry many assumptions, and I did not see those spelled out well.

This shouldn't be the only book in your bioinformatics library. If you really want algorithms, though, it's a good book to have in the collection and one you'll keep coming back to.

Rating:
Summary: Fantastic Descriptions of Probabilistic Sequence Algorithms
Review: I picked up this book at the recommendation of a number of colleagues in computational linguistics and speech processing as a way to find out what's going on in biological sequence analysis. I was hoping to learn about applications of the kinds of algorithms I know for handling speech and language, such as HMM decoding and context-free grammar parsing, to biological sequences. This book delivered, as recommended.

As the title implies, "Biological Sequence Analysis" focuses almost exlusively on sequence analysis. After a brief overview of statistics (more a reminder than an introduction), the first half of the book is devoted to alignment algorithms. These algorithms take pairs of sequences of bases making up DNA or sequences of amino acids making up proteins and provide optimal alignments of the sequences or of subsequences according to various statistical models of match likelihoods. Methods analyzed include edit distances with various substitution and gapping penalties (penalties for sections that don't match), Hidden Markov Models (HMMs) for alignment and also for classification against families, and finally, multiple sequence alignment, where alignment is generalized from pairs to sets of sequences. I found the section on building phylogenetic trees by means of hierarchical clustering to be the most fascinating section of the book (especially given its practical application to classifying wine varietals!). The remainder of the book is devoted to higher-order grammars such as context-free grammars, and their stochastic generalization. Stochastic context-free grammars are applied to the analysis of RNA secondary structure (folding). There is a good discussion of the CYK dynamic programming algorithm for non-deterministic context-free grammar parsing; an algorithm that is easily applied to finding the best parse in a probabilistic grammar. The presentations of the dynamic programming algorithms for HMM decoding, edit distance minimization, hierarchical clustering and context-free grammar parsing are as good as I've seen anywhere. They are precise, insightful, and informative without being overly subscripted. The illustrations provided are extremely helpful, including their positioning on pages where they're relevant.

This book is aimed at biologists trying to learn about algorithms, which is clear from the terse descriptions of the underlying biological problems. The technical details were so clear, though, that I was able to easily follow the algorithms even if I wasn't always sure about the genetic applications. After studying some introductions to genetics and coming back to this book, I was able to follow the application discussions much more easily. This book assumes the reader is familiar with algorithms and is comfortable manipulating a lot of statistics; a gentler introduction to exactly the same mathematics and algorithms can be found in Jurafsky and Martin's "Speech and Language Processing". For biologists who want to see how sequence statistics and algorithms applied to language, I would suggest Manning and Schuetze's "Foundations of Statistical Natural Language Processing". Although it is much more demanding computationally, more details on all of these algorithms, as well as some more background on the biology, along with some really nifty complexity analysis can be found in Dan Gusfield's "Algorithms on Strings, Trees and Sequences".

In these days of fly-by-night copy-editing and typesetting, I really appreciate Cambridge University Press's elegant style and attention to detail. Durbin, Eddy, Krogh and Mitchison's "Biological Sequence Analysis" is as beautiful and readable as it is useful.

Rating:
Summary: Good bargain, but...
Review: not suffciently precise for being an academic textbook. The definitions are sometimes incomplete, correctness proofs are missing, some exercises are incorrect. On the positive side, it does cover important topics, and brings good examples to illustrate main concepts and algorithms (which partially compemsates for the lack of precisenss).

Rating:
Summary: Good bargain, but...
Review: This book explained topics I was interested in above my personal expectations. All the mathematics and probabilistic models were explained in detail with a practical approach. I was even able to refine some of those models for specific needs without much previous experience nor knowledge. I highly recommend this book, it is one of the best I ever read.

Rating:
Summary: Excellent overview of probabilistic computational biology
Review: This book is a very well written overview to hidden Markov models and context-free grammar methods in computational biology. The authors have written a book that is useful to both biologists and mathematicians. Biologists with a background in probability theory equivalent to a senior-level course should be able to follow along without any trouble. The approach the author's take in the book is very intuitive and they motivate the concepts with elementary examples before moving on to the more abstract definitions. Exercises also abound in the book, and they are straightforward enough to work out, and should be if one desires an in-depth understanding of the main text. In addition, there is a software package called HMMER, developed by one of the authors (Eddy) that is in the public domain and can be downloaded from the Internet. The package specifically uses hidden Markov models to perform sequence analysis using the methods outlined in the book.

Probabilistic modeling has been applied to many different areas, including speech recognition, network performance analysis, and computational radiology. An overview of probabilistic modeling is given in the first chapter, and the authors effectively introduce the concepts without heavy abstract formalism, which for completeness they delegate to the last chapter of the book. Bayesian parameter estimation is introduced as well as maximum likelihood estimation. The authors take a pragmatic attitude in the utility of these different approaches, with both being developed in the book.

This is followed by a treatment of pairwise alignment in Chapter Two, which begins with substitution matrices. They point out, via some exercises, the role of physics in influencing particular alignments (hydrophobicity for example). Global alignment via the Gotoh algorithm and local alignment via the Smith-Waterman algorithm, are both discussed very effectively. Finite state machines with accompanying diagrams are used to discuss dynamic programming approaches to sequence alignment. The BLAST and FASTA packages are briefly discussed, along with the PAM and BLOSUM matrices.

Hidden Markov models are treated thoroughly in the next chapter with the Viterbi and Baum-Welch algorithms playing the central role. HIdden Markov models are then used in Chapter 4 for pairwise alignment. State diagrams are again used very effectively to illustrate the relevant ideas. Profile hidden Markov models which, according to the authors are the most popular application of hidden Markov models, are treated in detail in the next chapter. A very surprising application of Voronoi diagrams from computational geometry to weighting training sequences is given.

Several different approaches, such as Barton-Sternberg, CLUSTALW, Feng-Doolittle, MSA, simulated annealing, and Gibbs sampling are applied to multiple sequence alignment methods in Chapter 6. It is very well written, with the only disappointment being that only one exercise is given in the entire chapter. Phylogenetic trees are covered in Chapter 7, with emphasis placed on tree building algorithms using parsimony. The next chapter discusses the same topic from a probabilistic perspective. This to me was the most interesting part of the book as it connects the sequence alignment algorithms with evolutionary models.

The authors switch gears starting with the next chapter on transformational grammars. It is intriguing to see how concepts used in compiler construction can be generalized to the probabilistic case and then applied to computational biology. The PROSITE database is given as an example of the application of regular grammars to sequence matching. This chapter is fascinating reading, and there are some straightforward exercises illustrating the main points.

The last chapter covers RNA structure analysis, which introduces the concept of a pseudoknot. These are not to be confused with the usual knot constructions that can be applied to the topology of DNA, but instead result from the existence of non-nested base pairs in RNA sequences. The authors discuss many other techniques used in RNA sequence analysis and take care to point out which ones are more practical from a computational point of view. Surprisingly, genetic algorithms and algorithms based on Monte Carlo sampling are not discussed in the book, but the authors do give references for the interested reader.

The best attribute of this book is that the authors take a pragmatic point of view of how mathematics can be applied to problems in computational biology. They are not dogmatic about any particular approach, but instead fit the algorithm to the problem at hand.

Rating:
Summary: excellent, and tiny errors
Review: This is an excellent book, though no doubt challenging for some.(I lost two copies over the past three years, and recently bought a third.) Let me focus on some errors. Though tiny in comparison to the otherwise excellent content, it is helpful to point them out (especially since there is a paragraph devoted to each, in the book).

- Big-O notation on p. 21-22. Wrong. Big-O is not "of order". Big-O is asymptotic upper bound. Use Theta, when possible (see below). For this reason, you don't want to compare "O(n^3) with O(n^2)". Also note that not every algorithm has a Theta-bound; when not, you need separate analyses for worst-case, average-case, etc.

- "NP problems and intractability", p. 248-249. "NP problems are sometimes called intractable ...". Wrong. P is a subset of NP. NP-complete (or NP-hard, etc) problems are sometimes called intractable.

Rating:
Summary: Best practical introduction
Review: This is the best introduction to latest probabilistic sequence analysis methods. However, the book suffers from somewhat convoluted writing and organization. More importantly, it lacks a broader theoretical overview of the different methods. The methods are presented as a bunch of tools without enough critical assessment of their effectiveness or the relative strengths of their underlying theoretical models. I would have welcomed more discussion of how they all fit in a bigger probabilistic picture... what are the different simplifications and assumptions made for the sake of simplicity and computation?