General overview

Meetings are a rich resource of information that, in practice, is mostly untouched by any form of information processing. Even now it is rare that meetings are recorded, and fewer are then annotated for access purposes. Examples of the latter only include meetings held in parliaments, courts, hospitals, banks, etc., where a record is required for reasons of decision tracking or legal obligations. In these cases a labor-intensive manual transcription of the spoken words is produced. Giving much wider access to the rich content is the main aim of the AMI consortium projects, and there are now many examples of interest in that access – through the release of commercial hardware and software services. Especially with the advent of high-quality telephone and videoconferencing systems the opportunity to record, process, recognize, and categorize the interactions in meetings is recognized even by skeptics of speech and language processing technology.

Of course meetings are an audio-visual experience by nature and humans make extensive use of visual and other sensory information. To illustrate the rich landscape of information is the purpose of this book and many applications can be implemented even without looking at the spoken word. However, it is still verbal communication that forms the backbone of most meetings, and accounts for the bulk of the information transferred between participants. Hence automatic speech recognition (ASR) is key to access the information exchanged and is the most important part required for most higher level processing.

Methods of gene prediction

We saw in the previous chapter how prediction of CpG islands may be used to identify transcription start sites of protein-coding genes. However, there are many other elements and statistical properties of such genes that we may exploit for gene finding.

What are the methods available for computational gene finding? In general, one may distinguish between two major categories: de novo or ab initio methods and homology-based methods. The de novo methods make use of statistical signals in DNA sequences that are characteristic of protein-coding genes; the homology-based methods rely on the identification of exons by matching known mRNA or protein sequences or even profile HMMs to a genomic sequence. The homology-based methods are powerful but they require that mRNA or protein sequence information is available. Here we will focus on the de novo type of gene finding. What are the signals characteristic of proteincoding genes?

The RNA world

So far this book has focused on proteins and the genes that encode them. The human genome encodes some 21000 different proteins and the vast majority of them are important. On the other hand, there is a whole range of RNAs transcribed from the human genome that do not code for proteins, but have other functions. We refer to these RNAs as non-coding RNAs (ncRNAs). In fact, a major portion of the human genome is transcribed, although only about 1.5% of it corresponds to coding regions. We still do not know the function of many of these RNAs, but there are a large number of ncRNA families that have been characterized. Classic examples are tRNAs and ribosomal RNAs, which are part of the translation machinery. A set of U RNAs are involved in splicing (Chapter 16) and there are catalytically important RNA molecules of the RNA-processing enzymes RNases P and MRP. A vital and highly populated class of ncRNA is the RNAs involved in gene silencing as described in Chapter 3.

Meeting support technology evaluation can broadly be considered to be in three categories, which will be discussed in sequence in this chapter, in terms of goals, methods, and outcomes, following a brief introduction on methodology and undertakings prior to the AMI Consortium (Section 13.1). Evaluation efforts can be technology-centric, focused on determining how specific systems or interfaces performed in the tasks for which they were designed (Section 13.2). Evaluations can also adopt a task-centric view, defining common reference tasks such as fact finding or verification, which directly support cross-comparisons of different systems and interfaces (Section 13.3). Finally, the user-centric approach evaluates meeting support technology in its real context of use, measuring the increase in efficiency and user satisfaction that it brings (Section 13.4).

These aspects of evaluation differ from the component evaluation that accompanies each of the underlying technologies described in Chapters 3 to 10, which is often a black-box evaluation based on reference data and distance metrics (although task-centric approaches have been adopted for summarization evaluation, as shown in Chapter 10). Rather, the evaluation of meeting support technology is a stage in a complex software development process for which the helix model was proposed in Chapter 11. We think back on this process in the light of evaluation undertakings, especially for meeting browsers, at the end of this chapter (Section 13.5).

Approaches to evaluation: methods, experiments, campaigns

The evaluation of meeting browsers, as pieces of software, should be related (at least in theory) to a precise view of the specifications they answer.

This book has two parts: the first summarizes the facts of coding and information theory which are needed to understand the essence of estimation and statistics, and the second describes a new theory of estimation, which also covers a good part of statistics as well. After all, both estimation and statistics are about extracting information from the often chaotic looking data in order to learn what it is that makes the data behave the way they do. The first part together with an outline of the algorithmic information in Appendix A is meant for the statistician who wants to understand his or her discipline rather than just learn a bag of tricks with programs to apply them to various data, tricks that are not based on any theory and do not stand a critical examination although some of them can be quite useful, providing solutions for important statistical problems.

The word information has many meanings, two of which have been formalized by Shannon. The first is fundamental in communication as just the number of messages, strings of symbols, either to be stored or to be sent over some communication channel, the practical question being the size of the storage device needed or the time it takes to send them. The second meaning is the measure of the strength of the statistical property a string has, which is fundamental in statistics, and very different from that in communication.

For this chapter, as well as Chapters 12 and 13, we turn to the important genomics and bioinformatics problem of identifying biological function based on nucleotide and amino acid sequences.

Assigning function based on sequence similarity

A common problem in molecular biology is that you are faced with a gene or a gene product and you have no clue from experimental studies as to its function. In this context a critical contribution of bioinformatics is to attribute the sequence of a gene or a gene product a function. As one example, a genome sequencing project may give rise to tens of thousands of predicted protein sequences. In such a case we want to assign as many of these as possible a biological function using computational tools. In this manner we avoid many laborious wetlab experiments. In addition to genome sequencing projects, there are other more specialized situations where we want to find functions of genes. For instance, we could identify genes as being related to a specific genetic trait or disease, or a set of genes as being expressed under certain conditions.

A number of computational tools are available to predict a biological function associated with a protein sequence. In this chapter we will see an example in which we assign a function to a protein based on sequence similarity. Consider the human gene encoding the protein BRCA1, originally sequenced in 1994 (Miki et al., 1994). It was found to be related in sequence to a yeast protein RAD9. This yeast protein is involved in cell cycle control. This observation gave scientists a hint about possible roles of the BRCA1 gene. We see here an example of inferring a function based on a homology relationship to a protein that has already been functionally characterized. We will see yet another example of this situation in this chapter, where we will make use of BLAST to identify a homology relationship. We already encountered BLAST in the context of the BCR–ABL fusion protein in Chapter 7.

This chapter will deal further with phylogenetic analysis. We will introduce methods in addition to those of neighbour-joining and we will use a Perl script to examine taxonomy data. For these topics we will take a closer look at an extinct animal, the Tasmanian tiger.

Extinction

The Tasmanian tiger was not, in fact, a tiger; it was a dog-like marsupial animal. Thylacine is the more adequate scientific name. In the early twentieth century it existed only in Tasmania, and even there it was very scarce. A farmer named Wilf Batty lived in the Mawbanna district of northeastern Tasmania. On 13 May 1930 he spotted a thylacine attempting to break into his chicken coop. Batty had observed the thylacine around his house for weeks, and this day he took his rifle and shot the animal. As it happened, this was the last wild thylacine to be killed. Another specimen, most likely a female, was captured in 1933 and kept at the Hobart Zoo in Tasmania. She died on 7 September 1936, apparently as a result of neglect. The animal was kept outdoors and was not allowed access to her den, despite difficult temperatures. Ironically, the death took place only two months after the thylacine species was given full legal protection by the Tasmanian government. There are sightings of the thylacine reported after 1936, but none of these are well documented and we unfortunately need to regard the thylacine as being extinct.

Personal genomes

The first versions of the human genome sequence were presented in 2001 (Lander et al., 2001; Venter et al., 2001). They resulted from projects highly demanding in terms of resources and financing. The publicly funded sequencing project was supported by a $3 billion grant allocated to the Human Genome Project, and Craig Venter's project is reported to have cost $300 million. Since 2001, however, DNA sequencing technology has been made much more effective (Metzker, 2010) and the cost of sequencing a human genome has dropped dramatically. Now, in 2012, it is less than $5000 and is expected to become even less expensive. Furthermore, a human genome is now being sequenced in less than one week. Current DNA sequencing machines are able to produce gigabytes of data every day, and large sequencing centres are able to produce data at an unprecedented rate. For instance, the Beijing Genomics Institute (BGI) in Hong Kong is reported to have, as of December 2010, a total sequencing capacity of an astounding five terabases (5 × 1012, equivalent to 1000 human genomes) per day.

I have a long lasting interest in estimation, which started with attempts to control industrial processes. It did not take long to realize that the control part is easy if you knew the behavior of the process you want to control, which meant that the real problem is estimation. When I was asked by the Information Theory Society to give the 2009 Shannon Lecture I thought of giving a coherent survey of estimation theory. However, during the year given to prepare the talk I found that it was not possible, because there was no coherent theory of estimation. There was a collection of facts and results but they were isolated with little to connect them. To my surprise this applied even to the works of some of the greatest names in statistics, such as Fisher, Cramér, and Rao, which I had been familiar with for decades, but which I had never questioned until now that I was more or less forced to do so. As an example, the famous maximum likelihood estimator due to Fisher [12] had virtually no formal justification. The celebrated Cramér-Rao inequality gives it a non-asymptotic justification only for special models and for more general parametric models only an asymptotic justification. Clearly, no workable theory should be founded on asymptotic behavior. About the value of asymptotics, we quote Keynes' famous quip that “asymptotically we all shall be dead.”

In the last 30 years we have seen a dramatic development in molecular biology research. Genetic information has been mapped in great detail for many different living organisms. We are able to examine gene expression, biochemical reactions and molecular interactions within the cell in a manner that was quite impossible 50 years ago. This basic research has had a great impact on many areas, including medicine and biotechnology. For instance, molecular details of many diseases such as cancer have been worked out, making new methods of diagnosis and therapy possible. In addition, pharmaceutically important proteins such as insulin may be produced in high yield. In the world of plants, crops have been genetically modified to achieve increased crop yields and resistance to insects, or to make them produce specific substances in large quantities.

Extensive sugar decoration

In the two previous chapters we dealt with problems of assigning function to proteins or protein domains. We concerned ourselves with methods relying on sequence similarity, i.e. methods based on local alignment or profiles. We will now see an example of a functional domain which is not easily identified on the basis of sequence similarity or with profiles. It has specific properties as it has a characteristic amino acid composition and functional properties based on that composition. However, the properties of the domain may not be captured on the basis of sequence alignment or by position-specific information. The domain in question is characteristic of a group of proteins referred to as mucins (Perez-Vilar and Hill, 1999; Hollingsworth and Swanson, 2004).

A characteristic property of all mucins is the ability to form gels. Mucins are a major component of the mucous layer that is present on the surface of epithelial cells of the lung and intestine. The proteins act as a diffusion barrier to prevent harmful microorganisms and substances having more intimate contact with the cell. Mucins also function as lubricants to protect epithelial cells from dehydration and physical and chemical injury. Because mucins protect against pathogens, they play an important role in immune defence. In addition, certain mucins are associated with colon cancer (Hollingsworth and Swanson, 2004) and there is a strong association between the mucin Muc2 and the inflammatory bowel disease ulcerative colitis. For example, inflammation of the large intestine similar to ulcerative colitis is observed in mice that are deficient in the mucin Muc2 (Heazlewood et al., 2008).

Restriction enzymes, one of many tools of the molecular biology toolbox, were introduced in the previous chapter. The present chapter is devoted to another important method designed to examine the function of specific genes.

Interfering with gene expression

In order to study the biological function of a gene, the molecular biologist needs methods that allow alteration of that gene – for example, an alteration that reduces the production of the protein specified by the gene. The behaviour of a normal cell may then be compared to a cell where the gene of interest has been manipulated, thereby allowing a conclusion regarding the function of the gene of interest. In the yeast Saccharomyces cerevisiae, for instance, there are methods that allow the complete removal of genes (such a removal is often referred to as a gene knock-out ). If, for instance, the yeast gene named PSY3 is removed or inactivated, it gives rise to an increased level of mutations. This observation suggests to us that this particular gene is related to the repair of DNA damage.

For studies of human genes, you may want to examine a species more closely related to man than is yeast. You can delete or inactivate genes in mammals such as mice, but this is more technically involved than in yeast. On the other hand, there are other methods that make possible a reduction in gene expression in animals. One such important method involves RNA interference (RNAi) (also known as RNA silencing). In one common type of experiment, an mRNA is inactivated by the introduction of a small synthetic RNA which is complementary to the mRNA. In the bioinformatics example below we will see how we can design a small RNA for such an experiment. RNAi is an example of a gene knock-down method, as the expression of the target gene is reduced (compared to the gene knock-out, where a specific gene is completely removed or inactivated).

Elucidation of the human genome sequence was a significant milestone in the life sciences (Lander et al., 2001; Venter et al., 2001; International Human Genome Sequencing Consortium, 2004). However, with access to this information an obvious but entirely non-trivial problem was encountered. What does all the genetic information in the form of some three billion bases represent in biological terms? One important category of information is the sequences that specify genes, i.e. regions that give rise to mRNAs that in turn encode specific protein molecules. Not only proteins are specified by the genome; also a large number of RNAs are transcribed from DNA that do not give rise to mRNA, but have other functions. (These are non-coding RNAs and will be discussed in Chapter 17.) In the next three chapters we will deal with the computational problems of finding proteins and non-coding RNA genes, starting out with a genomic sequence.

When it comes to the protein-coding genes of a mammalian genome, only a very small fraction, about 1.5–2%, of the genome codes for protein. For these genomes we are faced with the problem of identifying relatively small and scattered coding regions in a vast sea of non-coding material. There is a striking difference in this respect between mammals and a bacterium like Escherichia coli, whose genome contains as much as 83% of coding sequence. In the next chapter the focus will be on prediction of exon regions of protein-coding genes. Here we will address another sub-problem of finding protein-coding genes. We will see how a simple prediction of regions known as CpG islands will help us to locate sites in the genome that are close to the transcription start sites of genes.