What is Bioinformatics? A Proposed Definition and Overview of the Field
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
346
© 2001 Schattauer GmbH
What is Bioinformatics?
A Proposed Definition and Overview of the Field
N. M. Luscombe, D. Greenbaum, M. Gerstein
Department of Molecular Biophysics and Biochemistry
Yale University, New Haven, USA
Summary 1. Introduction related to this new field has been surging,
and now comprise almost 2% of the
Background: The recent flood of data from genome annual total of papers in PubMed.
sequences and functional genomics has given rise to
Biological data are being produced at a This unexpected union between the two
new field, bioinformatics, which combines elements
of biology and computer science. phenomenal rate [1]. For example as subjects is attributed to the fact that life
Objectives: Here we propose a definition for this of April 2001, the GenBank repository of itself is an information technology; an
new field and review some of the research that is nucleic acid sequences contained organism’s physiology is largely deter-
being pursued, particularly in relation to transcriptional 11,546,000 entries [2] and the SWISS- mined by its genes, which at its most basic
regulatory systems. PROT database of protein sequences con- can be viewed as digital information.At the
Methods: Our definition is as follows: Bioinformatics tained 95,320 [3]. On average, these databa- same time, there have been major advances
is conceptualizing biology in terms of macromolecules ses are doubling in size every 15 months [2]. in the technologies that supply the initial
(in the sense of physical-chemistry) and then applying In addition, since the publication of data; Anthony Kervalage of Celera recent-
“informatics” techniques (derived from disciplines the H. influenzae genome [4], complete ly cited that an experimental laboratory
such as applied maths, computer science, and statis- sequences for nearly 300 organisms have can produce over 100 gigabytes of data a
tics) to understand and organize the information
been released, ranging from 450 genes to day with ease [5]. This incredible pro-
associated with these molecules, on a large-scale.
Results and Conclusions: Analyses in bioinformatics over 100,000. Add to this the data from the cessing power has been matched by devel-
predominantly focus on three types of large datasets myriad of related projects that study gene opments in computer technology; the most
available in molecular biology: macromolecular struc- expression, determine the protein structu- important areas of improvements have
tures, genome sequences, and the results of function- res encoded by the genes, and detail how been in the CPU, disk storage and Internet,
al genomics experiments (eg expression data). these products interact with one another, allowing faster computations, better data
Additional information includes the text of scientific and we can begin to imagine the enormous storage and revolutionalised the methods
papers and “relationship data” from metabolic path- quantity and variety of information that is for accessing and exchanging data.
ways, taxonomy trees, and protein-protein interaction being produced.
networks. Bioinformatics employs a wide range As a result of this surge in data, compu-
of computational techniques including sequence and ters have become indispensable to biologi-
structural alignment, database design and data
cal research. Such an approach is ideal
1.1 Aims of Bioinformatics
mining, macromolecular geometry, phylogenetic tree
construction, prediction of protein structure and because of the ease with which computers In general, the aims of bioinformatics are
function, gene finding, and expression data clustering. can handle large quantities of data and three-fold. First, at its simplest bioinfor-
The emphasis is on approaches integrating a variety of probe the complex dynamics observed in matics organises data in a way that allows
computational methods and heterogeneous data nature. Bioinformatics, the subject of the researchers to access existing information
sources. Finally, bioinformatics is a practical discipline. current review, is often defined as the appli- and to submit new entries as they are
We survey some representative applications, such as cation of computational techniques to produced, e.g. the Protein Data Bank for
finding homologues, designing drugs, and performing understand and organise the information 3D macromolecular structures [6, 7]. While
large-scale censuses. Additional information pertinent associated with biological macromolecules. data-curation is an essential task, the in-
to the review is available over the web at Fig. 1 shows that the number of papers formation stored in these databases is
http://bioinfo.mbb.yale.edu/what-is-it. essentially useless until analysed. Thus the
purpose of bioinformatics extends much
Keywords further. The second aim is to develop tools
Bioinformatics, Genomics, Introduction, Transcription Updated version of an invited review paper and resources that aid in the analysis of
Regulation that appeared in Haux, R., Kulikowski, C. (eds.)
(2001). IMIA Yearbook of Medical Informatics data. For example, having sequenced a par-
Method Inform Med 2001; 40: 346–58 2001: Digital Libraries and Medicine, pp. 83–99. ticular protein, it is of interest to compare it
Stuttgart: Schattauer. with previously characterised sequences.
Method Inform Med 4/2001347
What is Bioinformatics?
2001). At the next level are protein sequenc-
es comprising strings of 20 amino acid-
letters. At present there are about 400,000
known protein sequences [3], with a typical
bacterial protein containing approximately
300 amino acids. Macromolecular struc-
tural data represents a more complex form
of information. There are currently 15,000
entries in the Protein Data Bank, PDB
[6, 7], containing atomic structures of pro-
teins, DNA and RNA solved by x-ray
crystallography and NMR. A typical PDB
file for a medium-sized protein contains the
xyz-coordinates of approximately 2,000
atoms.
Scientific euphoria has recently centred
on whole genome sequencing. As with the
raw DNA sequences, genomes consist of
strings of base-letters, ranging from 1.6
million bases in Haemophilus influenzae
Fig. 1 Plot showing the growth of scientific publications in bioinformatics between 1973 and 2000. The histogram bars
(left vertical axis) counts the total number of scientific articles relating to bioinformatics, and the black line (right vertical [10] to 3 billion in humans [11, 12]. The
axis) gives the percentage of the annual total of articles relating to bioinformatics. The data are taken from PubMed. Entrez database [13] currently has com-
plete sequences for nearly 300 archaeal,
bacterial and eukaryotic organisms. In
This needs more than just a simple text- finally some of the major practical applica- addition to producing the raw nucleotide
based search, and programs such as FASTA tions of bioinformatics. sequence, a lot of work is involved in
[8] and PSI-BLAST [9] must consider what processing this data. An important aspect
constitutes a biologically significant match. of complete genomes is the distinction
Development of such resources dictates between coding regions and non-coding
expertise in computational theory, as well
as a thorough understanding of biology.
2. “…the INFORMATION regions -‘junk’ repetitive sequences making
up the bulk of base sequences especially in
The third aim is to use these tools to ana- associated with these eukaryotes. Within the coding regions,
lyse the data and interpret the results in a
biologically meaningful manner. Traditio- Molecules…” genes are annotated with their translated
protein sequence, and often with their
nally, biological studies examined individu- cellular function.
al systems in detail, and frequently com- Table 1 lists the types of data that are
pared them with a few that are related. In analysed in bioinformatics and the range of
bioinformatics, we can now conduct global topics that we consider to fall within the Bioinformatics – a Definition1
analyses of all the available data with the field. Here we take a broad view and in-
aim of uncovering common principles that clude subjects that may not normally be (Molecular) bio – informatics: bioinfor-
apply across many systems and highlight listed. We also give approximate values matics is conceptualising biology in
novel features. describing the sizes of data being discussed. terms of molecules (in the sense of Phy-
In this review, we provide a systematic We start with an overview of the sources sical chemistry) and applying “informa-
definition of bioinformatics as shown in of information. Most bioinformatics analy- tics techniques” (derived from disci-
Box 1. We focus on the first and third aims ses focus on three primary sources of data: plines such as applied maths, computer
just described, with particular reference to DNA or protein sequences, macromolecu- science and statistics) to understand and
the keywords: information, informatics, lar structures and the results of functional organise the information associated
organisation, understanding, large-scale genomics experiments. Raw DNA se- with these molecules, on a large scale. In
and practical applications. Specifically, we quences are strings of the four base-letters short, bioinformatics is a management
discuss the range of data that are currently comprising genes, each typically 1,000 bases information system for molecular biolo-
being examined, the databases into which long. The GenBank [2] repository of gy and has many practical applications.
they are organised, the types of analyses nucleic acid sequences currently holds a 1As submitted to the Oxford English
that are being conducted using transcrip- total of 12.5 billion bases in 11.5 million Dictionary.
tion regulatory systems as an example, and entries (all database figures as of April
Method Inform Med 4/2001348
Luscombe, Greenbaum, Gerstein
Table 1 Sources of data used in bioinformatics, the quantity of each type of data that is currently (April 2001) available, quences. While more biological informa-
and bioinformatics subject areas that utilize this data. tion can be derived from a single structure
than a protein sequence, the lack of depth
in the latter is compensated by analysing
larger quantities of data.
3. “… ORGANISE the Infor-
mation on a LARGE SCALE…”
3.1 Redundancy and Multiplicity
of Data
A concept that underpins most research
methods in bioinformatics is that much of
the data can be grouped together based on
biologically meaningful similarities. For
example, sequence segments are often
repeated at different positions of genomic
DNA [27]. Genes can be clustered into
those with particular functions (eg enzy-
matic actions) or according to the meta-
bolic pathway to which they belong [28],
although here, single genes may actually
possess several functions [29]. Going
further, distinct proteins frequently have
comparable sequences – organisms often
have multiple copies of a particular gene
through duplication and different species
have equivalent or similar proteins that
were inherited when they diverged from
each other in evolution. At a structural
level, we predict there to be a finite number
of different tertiary structures – estimates
More recent sources of data have been tities of data when experiments are con- range between 1,000 and 10,000 folds
from functional genomics experiments, of ducted for larger organisms and at more [30, 31] – and proteins adopt equivalent
which the most common are gene expres- time-points. structures even when they differ greatly in
sion studies. We can now determine expres- Further genomic-scale data include sequence [32]. As a result, although the
sion levels of almost every gene in a given biochemical information on metabolic number of structures in the PDB has
cell on a whole-genome level, however pathways, regulatory networks, protein- increased exponentially, the rate of discov-
there is currently no central repository for protein interaction data from two-hybrid ery of novel folds has actually decreased.
this data and public availability is limited. experiments, and systematic knockouts of There are common terms to describe the
These experiments measure the amount of individual genes to test the viability of an relationship between pairs of proteins or
mRNA that is produced by the cell [14-18] organism. the genes from which they are derived:
under different environmental conditions, What is apparent from this list is the analogous proteins have related folds, but
different stages of the cell cycle and differ- diversity in the size and complexity of dif- unrelated sequences, while homologous
ent cell types in multi-cellular organisms. ferent datasets. There are invariably more proteins are both sequentially and structu-
Much of the effort has so far focused on the sequence-based data than others because rally similar. The two categories can some-
yeast [19-24] and human genomes [25, 26]. of the relative ease with which they can be times be difficult to distinguish especially if
One of the largest dataset for yeast has produced.This is partly related to the great- the relationship between the two proteins
made approximately 20 time-point meas- er complexity and information-content of is remote [33, 34]. Among homologues, it is
urements for 6,000 genes [19]. However, individual structures or gene expression useful to distinguish between orthologues,
there is potential for much greater quan- experiments compared to individual se- proteins in different species that have evolv-
Method Inform Med 4/2001349
What is Bioinformatics?
ed from a common ancestral gene, and straightforward to access and cross- producing multiple sequence alignments
paralogues, proteins that are related by reference these sources of information be- [48], and searching for functional domains
gene duplication within a genome [35]. cause of differences in nomenclature and from conserved sequence motifs in such
Normally, orthologues retain the same file formats. alignments. Investigations of structural
function while paralogues evolve distinct, At a basic level, this problem is fre- data include prediction of secondary and
but related functions [36]. quently addressed by providing external tertiary protein structures, producing
An important concept that arises from links to other databases. For example in methods for 3D structural alignments [49,
these observations is that of a finite “parts PDBsum, web-pages for individual struc- 50], examining protein geometries using
list” for different organisms [37-39]: an tures direct the user towards corresponding distance and angular measurements, calcu-
inventory of proteins contained within an entries in the PDB, NDB, CATH, SCOP lations of surface and volume shapes and
organism, arranged according to different and SWISS-PROT databases. At a more analysis of protein interactions with other
properties such as gene sequence, protein advanced level, there have been efforts to subunits, DNA, RNA and smaller mole-
fold or function. Taking protein folds as an integrate access across several data sources. cules. These studies have lead to molecular
example, we mentioned that with a few One is the Sequence Retrieval System, SRS simulation topics in which structural data
exceptions, the tertiary structures of pro- [41], which allows flat-file databases to be are used to calculate the energetics in-
teins adopt one of a limited repertoire indexed to each other; this allows the user volved in stabilising macromolecular struc-
of folds. As the number of different fold to retrieve, link and access entries from tures, simulating movements within macro-
families is considerably smaller than the nucleic acid, protein sequence, protein molecules, and computing the energies
number of genes, categorising the proteins motif, protein structure and bibliographic involved in molecular docking. The increa-
by fold provides a substantial simplification databases. Another is the Entrez facility sing availability of annotated genomic
of the contents of a genome. Similar sim- [42], which provides similar gateways to sequences has resulted in the introduction
plifications can be provided by other attri- DNA and protein sequences, genome of computational genomics and proteomics
butes such as protein function. As such, we mapping data, 3D macromolecular structu- – large-scale analyses of complete genomes
expect this notion of a finite parts list to res and the PubMed bibliographic database and the proteins that they encode. Re-
become increasingly common in future [43].A search for a particular gene in either search includes characterisation of protein
genomic analyses. database will allow smooth transitions to content and metabolic pathways between
Clearly, an essential aspect of managing the genome it comes from, the protein different genomes, identification of interac-
this large volume of data lies in developing sequence it encodes, its structure, biblio- ting proteins, assignment and prediction of
methods for assessing similarities between graphic reference and equivalent entries for gene products, and large-scale analyses of
different biomolecules and identifying all related genes. In our own group, we have gene expression levels. Some of these re-
those that are related. There are well-docu- developed the SPINE [44] and PartsList search topics will be demonstrated in our
mented classifications for all of the main [39] web resources; these databases inte- example analysis of transcription regula-
types of data we described earlier. Al- grate many types of experimental data and tory systems.
though detailed descriptions of these clas- organise them using the concept of the Other subject areas we have included in
sification systems are beyond the scope of finite “parts list” we described above. Table 1 are: development of digital libraries
the current review, they are of great impor- for automated bibliographical searches,
tance as they ease comparisons between knowledge bases of biological information
genomes and their products. Links to the from the literature, DNA analysis methods
major databases are available from our
supplementary website.
4. “…UNDERSTAND and in forensics, prediction of nucleic acid struc-
tures, metabolic pathway simulations, and
Organise the Information…” linkage analysis – linking specific genes to
different disease traits.
Having examined the data, we can discuss In addition to finding relationships be-
3.2 Data Integration the types of analyses that are conducted.As tween different proteins, much of bioin-
The most profitable research in bioinfor- shown in Table 1, the broad subject areas in formatics involves the analysis of one type
matics often results from integrating mul- bioinformatics can be separated according of data to infer and understand the obser-
tiple sources of data [40]. For instance, the to the type of information that is used. For vations for another type of data. An exam-
3D coordinates of a protein are more useful raw DNA sequences, investigations involve ple is the use of sequence and structural
if combined with data about the protein’s separating coding and non-coding regions, data to predict the secondary and tertiary
function, occurrence in different genomes, and identification of introns, exons and structures of new protein sequences [51].
and interactions with other molecules. In promoter regions for annotating genomic These methods, especially the former, are
this way, individual pieces of information DNA [45, 46]. For protein sequences, ana- often based on statistical rules derived
are put in context with respect to other lyses include developing algorithms for from structures, such as the propensity for
data. Unfortunately, it is not always sequence comparisons [47], methods for certain amino acid sequences to produce
Method Inform Med 4/2001350
Luscombe, Greenbaum, Gerstein
Paradigm shifts during the past couple of decades have taken much of biology away from the
laboratory bench and have allowed the integration of other scientific disciplines, specifically
computing. The result is an expansion of biological research in breadth and depth. The vertical axis
demonstrates how bioinformatics can aid rational drug design with minimal work in the wet lab.
Starting with a single gene sequence, we can determine with strong certainty, the protein
sequence. From there, we can determine the structure using structure prediction techniques. With
geometry calculations, we can further resolve the protein’s surface and through molecular
simulation determine the force fields surrounding the molecule. Finally docking algorithms can
provide predictions of the ligands that will bind on the protein surface, thus paving the way for the
design of a drug specific to that molecule. The horizontal axis shows how the influx of biological
data and advances in computer technology have broadened the scope of biology. Initially with a pair
of proteins, we can make comparisons between the between sequences and structures of
evolutionary related proteins. With more data, algorithms for multiple alignments of several
proteins become necessary. Using multiple sequences, we can also create phylogenetic trees to
trace the evolutionary development of the proteins in question. Finally, with the deluge of data we
currently face, we need to construct large databases to store, view and deconstruct the
information. Alignments now become more complex, requiring sophisticated scoring schemes and
there is enough data to compile a genome census – a genomic equivalent of a population census –
providing comprehensive statistical accounting of protein features in genomes.
Fig. 2 Organizing and understanding biological data
different secondary structural elements. transferred between homologous proteins analysis in two dimension, depth and
Another example is the use of structural [55]. breadth. The first is represented by the
data to understand a protein’s function; vertical axis in the figure and outlines a
here studies have investigated the rela- possible approach to the rational drug
tionship different protein folds and their design process. The aim is to take a single
functions [52, 53] and analysed similarities
4.1 The Bioinformatics Spectrum gene and follow through an analysis that
between different binding sites in the ab- Fig. 2 summarises the main points we maximises our understanding of the
sence of homology [54]. Combined with raised in our discussions of organising protein it encodes. Starting with a gene
similarity measurements, these studies pro- and understanding biological data – the sequence, we can determine the protein
vide us with an understanding of how much development of bioinformatics techniques sequence with strong certainty. From there,
biological information can be accurately has allowed an expansion of biological prediction algorithms can be used to calcu-
Method Inform Med 4/2001351
What is Bioinformatics?
late the structure adopted by the protein.
Geometry calculations can define the
son methods such as text search and one-
dimensional alignment algorithms. Motif
6.1 Structural Studies
shape of the protein’s surface and molecu- and pattern identification for multiple As of April 2001, there were 379 structures
lar simulations can determine the force sequences depend on machine learning, of protein-DNA complexes in the PDB.
fields surrounding the molecule. Finally, clustering and data-mining techniques. 3D Analyses of these structures have provided
using docking algorithms, one could structural analysis techniques include Eu- valuable insight into the stereochemical
identify or design ligands that may bind clidean geometry calculations combined principles of binding, including how par-
the protein, paving the way for designing a with basic application of physical chemis- ticular base sequences are recognized
drug that specifically alters the protein’s try, graphical representations of surfaces and how the DNA structure is quite often
function. In practise, the intermediate steps and volumes, and structural comparison modified on binding.
are still difficult to achieve accurately, and and 3D matching methods. For molecular A structural taxonomy of DNA-binding
they are best combined with experimental simulations, Newtonian mechanics, quan- proteins, similar to that presented in SCOP
methods to obtain some of the data, for tum mechanics, molecular mechanics and and CATH, was first proposed by Harrison
example characterising the structure of the electrostatic calculations are applied. In [56] and periodically updated to accom-
protein of interest. many of these areas, the computational modate new structures as they are solved
The aim of the second dimension, the methods must be combined with good [57]. The classification consists of a two-tier
breadth in biological analysis, is to compare statistical analyses in order to provide an system: the first level collects proteins into
a gene or gene product with others. Ini- objective measure for the significance of eight groups that share gross structural
tially, simple algorithms can be used to the results. features for DNA-binding, and the second
compare the sequences and structures of a comprises 54 families of proteins that are
pair of related proteins. With a larger num- structurally homologous to each other.
ber of proteins, improved algorithms can be
used to produce multiple alignments, and
6. Transcription Regulation – Assembly of such a system simplifies the
comparison of different binding methods; it
extract sequence patterns or structural a Case Study in Bioinformatics highlights the diversity of protein-DNA
templates that define a family of proteins. complex geometries found in nature, but
Using this data, it is also possible to con- DNA-binding proteins have a central role also underlines the importance of inter-
struct phylogenetic trees to trace the evolu- in all aspects of genetic activity within an actions between -helices and the DNA
tionary path of proteins. Finally, with even organism, participating in processes such as major groove, the main mode of binding in
more data, the information must be stored transcription, packaging, rearrangement, over half the protein families. While the
in large-scale databases. Comparisons replication and repair. In this section, we number of structures represented in the
become more complex, requiring multiple focus on the studies that have contributed PDB does not necessarily reflect the rela-
scoring schemes, and we are able to con- to our understanding of transcription tive importance of the different proteins in
duct genomic scale censuses that provide regulation in different organisms. Through the cell, it is clear that helix-turn-helix,
comprehensive statistical accounts of this example, we demonstrate how bio- zinc-coordinating and leucine zipper motifs
protein features, such as the abundance of informatics has been used to increase our are used repeatedly.These provide compact
particular structures or functions in diffe- knowledge of biological systems and also frameworks to present the -helix on the
rent genomes. It also allows us to build illustrate the practical applications of the surfaces of structurally diverse proteins. At
phylogenetic trees that trace the evolution different subject areas that were briefly a gross level, it is possible to highlight the
of whole organisms. outlined earlier. We start by considering differences between transcription factor
structural analyses of how DNA-binding domains that “just” bind DNA and those
proteins recognise particular base se- involved in catalysis [58]. Although there
5. “… applying INFORMATICS quences. Later, we review several genomic
studies that have characterised the nature
are exceptions, the former typically
approach the DNA from a single face and
TECHNIQUES…” of transcription factors in different orga- slot into the grooves to interact with base
nisms, and the methods that have been used edges. The latter commonly envelope the
The distinct subject areas we mention to identify regulatory binding sites in the substrate, using complex networks of
require different types of informatics tech- upstream regions. Finally, we provide an secondary structures and loops.
niques. Briefly, for data organisation, the overview of gene expression analyses that Focusing on proteins with -helices, the
first biological databases were simple flat have been recently conducted and suggest structures show many variations, both in
files. However with the increasing amount future uses of transcription regulatory ana- amino acid sequences and detailed geo-
of information, relational database lyses to rationalise the observations made metry. They have clearly evolved indepen-
methods with Web-page interfaces have in gene expression experiments. All the dently in accordance with the requirements
become increasingly popular. In sequence results that we describe have been found of the context in which they are found.
analysis, techniques include string compari- through computational studies. While achieving a close fit between the
Method Inform Med 4/2001352 Luscombe, Greenbaum, Gerstein -helix and major groove, there is enough interactions of arginine or lysine with 6.2 Genomic Studies flexibility to allow both the protein and guanine, asparagine or glutamine with DNA to adopt distinct conformations. adenine and threonine with thymine. Such Due to the wealth of biochemical data that However, several studies that analysed the preferences were explained through exami- are available, genomic studies in bioin- binding geometries of -helices demon- nation of the stereochemistry of the amino formatics have concentrated on model strated that most adopt fairly uniform con- acid side chains and base edges. Also organisms, and the analysis of regulatory formations regardless of protein family. highlighted were more complex types of systems has been no exception. Identification They are commonly inserted in the major interactions where single amino acids of transcription factors in genomes invari- groove sideways, with their lengthwise axis contact more than one base-step simulta- ably depends on similarity search strate- roughly parallel to the slope outlined by neously, thus recognising a short DNA gies, which assume a functional and evolu- the DNA backbone. Most start with the sequence. These results suggested that tionary relationship between homologous N-terminus in the groove and extend out, universal specificity, one that is observed proteins. In E. coli, studies have so far completing two to three turns within across all protein-DNA complexes, indeed estimated a total of 300 to 500 transcription contacting distance of the nucleic acid [59, exists. However, many interactions that are regulators [71] and PEDANT [72], a data- 60]. normally considered to be non-specific, base of automatically assigned gene funct- Given the similar binding orientations, it such as those with the DNA backbone, can ions, shows that typically 2-3% of pro- is surprising to find that the interactions also provide specificity depending on the karyotic and 6-7% of eukaryotic genomes between each amino acid position along context in which they are made. comprise DNA-binding proteins.As assign- the -helices and nucleotides on the DNA Armed with an understanding of ments were only complete for 40-60% of vary considerably between different pro- protein structure, DNA-binding motifs and genomes as of August 2000, these figures tein families. However, by classifying the side chain stereochemistry, a major applica- most likely underestimate the actual num- amino acids according to the sizes of their tion has been the prediction of binding ber. Nonetheless, they already represent a side chains, we are able to rationalise the either by proteins known to contain a parti- large quantity of proteins and it is clear that different interactions patterns. The rules of cular motif, or those with structures solved there are more transcription regulators interactions are based on the simple pre- in the uncomplexed form. Most common in eukaryotes than other species. This is mise that for a given residue position on are predictions for -helix-major groove unsurprising, considering the organisms -helices in similar conformations, small interactions – given the amino acid se- have developed a relatively sophisticated amino acids interact with nucleotides that quence, what DNA sequence would it transcription mechanism. are close in distance and large amino acids recognise [61, 67]. In a different approach, From the conclusions of the structural with those that are further [60, 61]. Equi-va- molecular simulation techniques have been studies, the best strategy for characterising lent studies for binding by other structural used to dock whole proteins and DNAs on DNA-binding of the putative transcription motifs, like -hairpins, have also been con- the basis of force-field calculations around factors in each genome is to group them ducted [62]. When considering these the two molecules [68, 69]. by homology and to analyse the individual interactions, it is important to remember The reason that both methods have families. Such classifications are provided that different regions of the protein surface been met with limited success is because in the secondary sequence databases also provide interfaces with the DNA. even for apparently simple cases like - described earlier and also those that This brings us to look at the atomic level helix-binding, there are many other factors specialise in regulatory proteins such as interactions between individual amino that must be considered. Comparisons RegulonDB [73] and TRANSFAC [74]. acid-base pairs. Such analyses are based on between bound and unbound nucleic acid Of even greater use is the provision of the premise that a significant proportion of structures show that DNA-bending is a structural assignments to the proteins; specific DNA-binding could be rationalised common feature of complexes formed with given a transcription factor, it is helpful to by a universal code of recognition between transcription factors [58, 70].This and other know the structural motif that it uses for amino acids and bases, ie whether certain factors such as electrostatic and cation- binding, therefore providing us with a protein residues preferably interact with mediated interactions assist indirect better understanding of how it recognises particular nucleotides regardless of the recognition of the nucleotide sequence, the target sequence. Structural genomics type of protein-DNA complex [63]. Studies although they are not well understood yet. through bioinformatics assigns structures have considered hydrogen bonds, van der Therefore, it is now clear that detailed rules to the protein products of genomes by Waals contacts and water-mediated bonds for specific DNA-binding will be family demonstrating similarity to proteins of [64-66]. Results showed that about 2/3 of all specific, but with underlying trends such as known structure [75]. These studies have interactions are with the DNA back- the arginine-guanine interactions. shown that prokaryotic transcription fac- bone and that their main role is one of tors most frequently contain helix-turn- sequence-independent stabilisation. In helix motifs [71, 76] and eukaryotic factors contrast, interactions with bases display contain homeodomain type helix-turn- some strong preferences, including the helix, zinc finger or leucine zipper motifs. Method Inform Med 4/2001
353
What is Bioinformatics?
From the protein classifications in each specific but different members bind distinct approach, it was found that at least 27% of
genome, it is clear that different types of base sequences. Here protein residues known E. coli DNA-regulatory motifs are
regulatory proteins differ in abundance and undergo frequent mutations, and family conserved in one or more distantly related
families significantly differ in size. A study members can be divided into subfamilies bacteria [84].
by Huynen and van Nimwegen [77] has according to the amino acid sequences The detection of regulatory sites in
shown that members of a single family have at base-contacting positions; those in the eukaryotes poses a more difficult problem
similar functions, but as the requirements same subfamily are predicted to bind the because consensus sequences tend to be
of this function vary over time, so does same DNA sequence and those of different much shorter, variable, and dispersed over
the presence of each gene family in the subfamilies to bind distinct sequences. On very large distances. However, initial stud-
genome. the whole, the subfamilies corresponded ies in S. cerevisiae provided an interesting
Most recently, using a combination of well with the proteins’ functions and mem- observation for the GATA protein in nitro-
sequence and structural data, we examined bers of the same subfamilies were found to gen metabolism regulation. While the 5
the conservation of amino acid sequences regulate similar transcription pathways. base-pair GATA consensus sequence is
between related DNA-binding proteins, The combined analysis of sequence and found almost everywhere in the genome, a
and the effect that mutations have on structural data described by this study pro- single isolated binding site is insufficient to
DNA sequence recognition. The structural vided an insight into how homologous exert the regulatory function [85]. There-
families described above were expanded to DNA-binding scaffolds achieve different fore specificity of GATA activity comes
include proteins that are related by sequence specificities by altering their amino acid from the repetition of the consensus se-
similarity, but whose structures remain sequences. In doing so, proteins evolved quence within the upstream regions of con-
unsolved. Again, members of the same distinct functions, therefore allowing trolled genes in multiple copies. An initial
family are homologous, and probably derive structurally related transcription factors to study has used this observation to predict
from a common ancestor. regulate expression of different genes. new regulatory sites by searching for over-
Amino acid conservations were calculat- Therefore, the relative abundance of tran- represented oligonucleotides in non-coding
ed for the multiple sequence alignments scription regulatory families in a genome regions of yeast and worm genomes [86,
of each family [78]. Generally, alignment depends, not only on the importance of a 87].
positions that interact with the DNA are particular protein function, but also in the Having detected the regulatory binding
better conserved than the rest of the pro- adaptability of the DNA-binding motifs to sites, there is the problem of defining the
tein surface, although the detailed patterns recognise distinct nucleotide sequences. genes that are actually regulated, commonly
of conservation are quite complex. Residues This, in turn, appears to be best accommo- termed regulons. Generally, binding sites
that contact the DNA backbone are highly dated by simple binding motifs, such as the are assumed to be located directly upstream
conserved in all protein families, providing zinc fingers. of the regulons; however there are different
a set of stabilising interactions that are Given the knowledge of the transcription problems associated with this assumption
common to all homologous proteins. The regulators that are contained in each depending on the organism. For prokary-
conservation of alignment positions that organism, and an understanding of how otes, it is complicated by the presence of
contact bases, and recognise the DNA they recognise DNA sequences, it is of operons; it is difficult to locate the regulat-
sequence, are more complex and could be interest to search for their potential bind- ed gene within an operon since it can lie
rationalised by defining a three-class model ing sites within genome sequences [79]. several genes downstream of the regulatory
for DNA-binding. First, protein families For prokaryotes, most analyses have in- sequence. It is often difficult to predict the
that bind non-specifically usually contain volved compiling data on experimentally organisation of operons [88], especially to
several conserved base-contacting residues; known binding sites for particular proteins define the gene that is found at the head,
without exception, interactions are made in and building a consensus sequence that in- and there is often a lack of long-range con-
the minor groove where there is little corporates any variations in nucleotides. servation in gene order between related
discrimination between base types. The Additional sites are found by conducting organisms [89]. The problem in eukaryotes
contacts are commonly used to stabilise word-matching searches over the entire is even more severe; regulatory sites often
deformations in the nucleic acid structure, genome and scoring candidate sites by act in both directions, binding sites are
particularly in widening the DNA minor similarity [80-83]. Unsurprisingly, most of usually distant from regulons because of
groove. The second class comprise families the predicted sites are found in non-coding large intergenic regions, and transcription
whose members all target the same regions of the DNA [80] and the results of regulation is usually a result of combined
nucleotide sequence; here, base-contacting the studies are often presented in databases action by multiple transcription factors in a
positions are absolutely or highly conser- such as RegulonDB [73]. The consensus combinatorial manner.
ved allowing related proteins to target the search approach is often complemented by Despite these problems, these studies
same sequence. comparative genomic studies searching have succeeded in confirming the transcrip-
The third, and most interesting, class upstream regions of orthologous genes in tion regulatory pathways of well-character-
comprises families in which binding is also closely related organisms. Through such an ized systems such as the heat shock response
Method Inform Med 4/2001354
Luscombe, Greenbaum, Gerstein
system [83]. In addition, it is feasible to relates to the functions associated with the distinction between different forms of
experimentally verify any predictions, most these folds; the former are commonly in- cancer – for example subclasses of acute
notably using gene expression data. volved in metabolic pathways and the latter leukaemia – has been well established, it is
in signalling or transport processes [98]. still not possible to establish a clinical diag-
This is also reflected in the relationship nosis on the basis of a single test. In a
with subcellular localisations of proteins, recent study, acute myeloid leukaemia and
6.3 Gene Expression Studies where expression of cytoplasmic proteins is acute lymphoblastic leukaemia were suc-
Many expression studies have so far high, but nuclear and membrane proteins cessfully distinguished based on the ex-
focused on devising methods to cluster tend to be low [99, 100]. pression profiles of these cells [26]. As the
genes by similarities in expression profiles. More complex relationships have also approach does not require prior biological
This is in order to determine the proteins been assessed. Conventional wisdom is that knowledge of the diseases, it may provide a
that are expressed together under different gene products that interact with each other generic strategy for classifying all types of
cellular conditions. Briefly, the most com- are more likely to have similar expression cancer.
mon methods are hierarchical clustering, profiles than if they do not [101, 102]. How- Clearly, an essential aspect of under-
self-organising maps, and K-means cluster- ever, a recent study showed that this rela- standing expression data lies in understand-
ing. Hierarchical methods originally de- tionship is not so simple [103]. While ex- ing the basis of transcription regulation.
rived from algorithms to construct phyloge- pression profiles are similar for gene pro- However, analysis in this area is still limited
netic trees, and group genes in a “bottom- ducts that are permanently associated, for to preliminary analyses of expression levels
up” fashion; genes with the most similar example in the large ribosomal subunit, in yeast mutants lacking key components of
expression profiles are clustered first, and profiles differ significantly for products the transcription initiation complex [19,
those with more diverse profiles are that are only associated transiently, includ- 107].
included iteratively [90-92]. In contrast, the ing those belonging to the same metabolic
self-organising map [93, 94] and K-means pathway.
methods [95, 96] employ a “top-down”
approach in which the user pre-defines the
As described below, one of the main
driving forces behind expression analysis
7 “… many PRACTICAL
number of clusters for the dataset. The has been to analyse cancerous cell lines APPLICATIONS…”
clusters are initially assigned randomly, and [104]. In general, it has been shown that dif-
the genes are regrouped iteratively until ferent cell lines (eg epithelial and ovarian Here, we describe some of the major uses
they are optimally clustered. cells) can be distinguished on the basis of of bioinformatics.
Given these methods, it is of interest to their expression profiles, and that these
relate the expression data to other attri- profiles are maintained when cells are 7.1 Finding Homologues
butes such as structure, function and sub- transferred from an in vivo to an in vitro
cellular localisation of each gene product. environment [105]. The basis for their phy- As described earlier, one of the driving
Mapping these properties provides an siological differences were apparent in the forces behind bioinformatics is the search
insight into the characteristics of proteins expression of specific genes; for example, for similarities between different biomole-
that are expressed together, and also sug- expression levels of gene products neces- cules. Apart from enabling systematic orga-
gest some interesting conclusions about the sary for progression through the cell cycle, nisation of data, identification of protein
overall biochemistry of the cell. In yeast, especially ribosomal genes, correlated well homologues has some direct practical uses.
shorter proteins tend to be more highly with variations in cell proliferation rate. The most obvious is transferring informa-
expressed than longer proteins, probably Comparative analysis can be extended to tion between related proteins. For example,
because of the relative ease with which they tumour cells, in which the underlying given a poorly characterised protein, it is
are produced [97]. Looking at the amino causes of cancer can be uncovered by possible to search for homologues that are
acid content, highly expressed genes are pinpointing areas of biological variations better understood and with caution, apply
generally enriched in alanine and glycine, compared to normal cells. For example in some of the knowledge of the latter to the
and depleted in asparagine; these are breast cancer, genes related to cell prolif- former. Specifically with structural data,
thought to reflect the requirements of eration and the IFN-regulated signal trans- theoretical models of proteins are usually
amino acid usage in the organism, where duction pathway were found to be upregu- based on experimentally solved structures
synthesis of alanine and glycine are energe- lated [25, 106]. One of the difficulties in of close homologues [108]. Similar tech-
tically less expensive than asparagine.Turn- cancer treatment has been to target specific niques are used in fold recognition in which
ing to protein structure, expression levels of therapies to pathogenetically distinct tu- tertiary structure predictions depend on
the TIM barrel and NTP hydrolase folds mour types, in order to maximise efficacy finding structures of remote homologues
are highest, while those for the leucine and minimise toxicity. Thus, improvements and checking whether the prediction is
zipper, zinc finger and transmembrane in cancer classifications have been central energetically viable [109]. Where biochem-
helix-containing folds are lowest. This to advances in cancer treatment. Although ical or structural data are lacking, studies
Method Inform Med 4/2001355
What is Bioinformatics?
Fig. 3 Above is a schematic outlining how scientists can use bioinformatics to aid rational drug discovery. MLH1 is a human gene encoding a mismatch repair protein (mmr) situated on the
short arm of chromosome 3. Through linkage analysis and its similarity to mmr genes in mice, the gene has been implicated in nonpolyposis colorectal cancer. Given the nucleotide sequence,
the probable amino acid sequence of the encoded protein can be determined using translation software. Sequence search techniques can be used to find homologues in model organisms, and
based on sequence similarity, it is possible to model the structure of the human protein on experimentally characterised structures. Finally, docking algorithms could design molecules that
could bind the model structure, leading the way for biochemical assays to test their biological activity on the actual protein.
could be made in low-level organisms like simplifies the problem of understanding missing in humans. On a smaller scale,
yeast and the results applied to homo- complex genomes by analysing simple structural differences between similar pro-
logues in higher-level organisms such as organisms first and then applying the teins may be harnessed to design drug
humans, where experiments are more same principles to more complicated ones – molecules that specifically bind to one
demanding. this is one reason why early structural structure but not another.
An equivalent approach is also employed genomics projects focused on Mycoplasma
in genomics. Homologue-finding is exten- genitalium [75].
sively used to confirm coding regions in Ironically, the same idea can be applied
newly sequenced genomes and functional in reverse. Potential drug targets are quickly
7.2 Rational Drug Design
data is frequently transferred to annotate discovered by checking whether homo- One of the earliest medical applications of
individual genes. On a larger scale, it also logues of essential microbial proteins are bioinformatics has been in aiding rational
Method Inform Med 4/2001356
Luscombe, Greenbaum, Gerstein
drug design. Fig. 3 outlines the commonly protein folds are often related to specific this way, we are able to examine individual
cited approach, taking the MLH1 gene pro- biochemical functions [52, 53], these find- systems in detail and also compare them
duct as an example drug target. MLH1 is a ings highlight the diversity of metabolic with those that are related in order to un-
human gene encoding a mismatch repair pathways in different organisms [36, 89]. cover common principles that apply across
protein (mmr) situated on the short arm of As we discussed earlier, one of the most many systems and highlight unusual fea-
chromosome 3 [110]. Through linkage ana- exciting new sources of genomic information tures that are unique to some.
lysis and its similarity to mmr genes in mice, is the expression data. Combining expression
the gene has been implicated in nonpoly- information with structural and functional Acknowledgments
We thank Patrick McGarvey for comments on
posis colorectal cancer [111]. Given the classifications of proteins we can ask the manuscript.
nucleotide sequence, the probable amino whether the high occurrence of a protein
acid sequence of the encoded protein can fold in a genome is indicative of high ex-
be determined using translation software. pression levels [97]. Further genomic scale
Sequence search techniques can then be data that we can consider in large-scale sur-
References
1. Reichhardt T. It’s sink or swim as a tidal wave
used to find homologues in model orga- veys include the subcellular localisations of of data approaches. Nature 1999. 399 (6736):
nisms, and based on sequence similarity, it proteins and their interactions with each 517-20.
is possible to model the structure of the other [114-116]. In conjunction with struc- 2. Benson DA, et al. GenBank. Nucleic Acids
human protein on experimentally character- tural data, we can then begin to compile a Res 2000; 28 (1): 15-8.
3. Bairoch A, Apweiler R. The SWISS-PROT
ized structures. Finally, docking algorithms map of all protein-protein interactions in protein sequence database and its supplement
could design molecules that could bind the an organism. TrEMBL in 2000. Nucleic Acids Res 2000; 28
model structure, leading the way for bio- (1): 45-8.
chemical assays to test their biological 4. Fleischmann RD, et al. Whole-genome ran-
dom sequencing and assembly of Haemo-
activity on the actual protein. philus influenzae Rd. Science 1995; 269
8. Conclusions (5223): 496-512.
5. Drowning in data. The Economist 1999 (26
With the current deluge of data, compu- June 1999).
7.3 Large-scale Censuses tational methods have become indispens-
6. Bernstein FC, et al. The Protein Data Bank. A
computer-based archival file for macromolec-
Although databases can efficiently store all able to biological investigations. Originally ular structures. Eur J Biochem 1977; 80 (2):
the information related to genomes, struc- developed for the analysis of biological se- 319-24.
tures and expression datasets, it is useful to quences, bioinformatics now encompasses 7. Berman HM, et al. The Protein Data Bank.
Nucleic Acids Res 2000; 28 (1): 235-42.
condense all this information into under- a wide range of subject areas including 8. Pearson WR, Lipman DJ. Improved tools for
standable trends and facts that users can structural biology, genomics and gene ex- biological sequence comparison. Proc Natl
readily understand. Broad generalisations pression studies. In this review, we provided Acad Sci USA 1988; 85 (8): 2444-8.
9. Altschul SF, et al. Gapped BLAST and PSI-
help identify interesting subject areas for an introduction and overview of the cur- BLAST: a new generation of protein database
further detailed analysis, and place new ob- rent state of field. In particular, we discus- search programs. Nucleic Acids Res 1997; 25
servations in a proper context. This enables sed the types of biological information and (17): 3389-402.
one to see whether they are unusual in any databases that are commonly used, exa- 10. Fleischmann RD, et al. Whole-genome ran-
dom sequencing and assembly of Haemo-
way. mined some of the studies that are being con- philus influenzae Rd. Science 1995; 269
Through these large-scale censuses, one ducted – with reference to transcription (5223): 496-512.
can address a number of evolutionary, bio- regulatory systems – and finally looked at 11. Lander ES, et al. Initial sequencing and analy-
chemical and biophysical questions. For several practical applications of the field. sis of the human genome. Nature 2001; 409:
860-921.
example, are specific protein folds associat- Two principal approaches underpin all stud- 12. Venter JC, et al. The sequence of the human
ed with certain phylogenetic groups? How ies in bioinformatics. First is that of com- genome. Science 2001; 291 (5507): 1304-51.
common are different folds within partic- paring and grouping the data according to 13. Tatusova TA, Karsch-Mizrachi I, Ostell JA.
ular organisms? And to what degree are biologically meaningful similarities and sec- Complete genomes in WWW Entrez: data
representation and analysis. Bioinformatics
folds shared between related organisms? ond, that of analysing one type of data to 1999; 15 (7-8): 536-43.
Does this extent of sharing parallel meas- infer and understand the observations for 14. Eisen MB, Brown PO. DNA arrays for analy-
ures of relatedness derived from traditional another type of data. These approaches are sis of gene expression. Methods Enzymol,
evolutionary trees? Initial studies show reflected in the main aims of the field, 1999; 303: 179-205.
15. Cheung VG, et al. Making and reading micro-
that the frequency of folds differs greatly which are to understand and organise the arrays. Nat Genet 1999; 21 (1 Suppl): 15-9.
between organisms and that the sharing of information associated with biological 16. Duggan DJ, et al. Expression profiling using
folds between organisms does in fact follow molecules on a large scale. As a result, cDNA microarrays. Nat Genet 1999. 21
(1 Suppl): 10-4.
traditional phylogenetic classifications bioinformatics has not only provided great-
17. Lipshutz RJ, et al. High density synthetic
[37, 112, 113]. We can also integrate data on er depth to biological investigations, but oligonucleotide arrays. Nat Genet 1999; 21 (1):
protein functions; given that the particular added the dimension of breadth as well. In 20-4.
Method Inform Med 4/2001357
What is Bioinformatics?
18. Velculescu VE, et al. Serial Analysis of Gene 37. Gerstein M, Hegyi H. Comparing genomes in sequence, structure and function through
Expression. Detailed Protocol 1999. terms of protein structure: surveys of a finite traditional and probabilistic scores. J Mol Biol
19. Holstege FC, et al. Dissecting the regulatory parts list. FEMS Microbiol Rev 1998; 22 (4): 2000; 297 (1): 233-49.
circuitry of a eukaryotic genome. Cell 1998; 95 277-304. 56. Harrison SC. A structural taxonomy of DNA-
(5): 717-28. 38. Skolnick J, Fetrow JS. From genes to protein binding domains. Nature 1991; 353 (6346):
20. Roth FP, Estep PW, Church GM. Finding structure and function: novel applications of 715-9.
DNA regulatory motifs within unaligned non- computational approaches in the genomic era. 57. Luscombe NM, et al.An overview of the struc-
coding sequences clustered by whole-genome Trends Biotech 2000; 18: 34-9. tures of protein-DNA complexes. Genome
mRNA quantitation. Nat Biotech 1998; 16 39. Qian J, et al. PartsList: a web-based system for Biology 2000; 1 (1): 1-37.
(10): 939-45. dynamically ranking protein folds based on 58. Jones S, et al. Protein-DNA interactions: A
21. Jelinsky SA, Samson LD. Global response of disparate attributes, including whole-genome structural analysis. J Mol Biol 1999; 287 (5):
Saccharomyces cerevisiae to an alkylating expression and interaction information. 877-96.
agent. Proc Natl Acad Sci USA 1999; 96 (4): Nucleic Acids Res 2001; 29 (8): 1750-64. 59. Suzuki M, Gerstein M. Binding geometry of
1486-91. 40. Gerstein M. Integrative database analysis in alpha-helices that recognize DNA. Proteins
structural genomics. Nat Struct Biol 2000; 7 1995; 23 (4): 525-35.
22. Cho RJ, et al. A genome-wide transcriptional Suppl: 960-3.
analysis of the mitotic cell cycle. Mol Cell 60. Luscombe NM, Thornton JM. Protein-DNA
41. Etzold T, Ulyanov A, Argos P. SRS: informa- interactions: a 3D analysis of alpha-helix-
1998; 2 (1): 65-73. tion retrieval system for molecular biology data binding in the major groove. Manuscript in
23. DeRisi JL, Iyer VR, Brown PO. Exploring the banks. Methods Enzymol 1996; 266: 114-28. preparation.
metabolic and genetic control of gene expres- 42. Schuler GD, et al. Entrez: molecular biology
sion on a genomic scale. Science 1997; 278 61. Suzuki M, et al. DNA recognition code of
database and retrieval system. Methods transcription factors. Protein Eng 1995; 8 (4):
(5338): 680-6. Enzymol 1996; 266: 141-62. 319-28.
24. Winzeler EA, et al. Functional characteriza- 43. Wade K. Searching Entrez PubMed and 62. Suzuki M. DNA recognition by a -sheet.
tion of the S. cerevisiae genome by gene uncover on the internet. Aviat Space Environ Protein Eng 1995; 8 (1): 1-4.
deletion and parallel analysis. Science 1999; Med 2000; 71 (5): 559.
63. Seeman NC, Rosenberg JM, Rich A. Sequence
285 (5429): 901-6. 44. Bertone P, et al. SPINE: An integrated
specific recognition of double helical nucleic
25. Perou CM, et al. Molecular portraits of human tracking database and datamining approach
acids by proteins. Proc Natl Acad Sci USA
breast tumours. Nature 2000; 406 (6797): for high-throughput structural proteomics,
1976; 73: 804-8.
747-52. enabling the determination of the properties
64. Suzuki M. A framework for the DNA-protein
26. Golub TR, et al. Molecular classification of of readily characterized proteins. Nucleic
recognition code of the probe helix in
cancer: class discovery and class prediction by Acids Res. In Press.
transcription factors: the chemical and stereo-
gene expression monitoring. Science 1999; 286 45. Zhang MQ. Promoter analysis of co-regulated
chemical rules. Structure 1994; 2 (4): 317-26.
(5439): 531-7. genes in the yeast genome. Comput Chem
1999; 23 (3-4): 233-50. 65. Mandel-Gutfreund Y, Schueler O, Margalit H.
27. Pedersendagger AG, et al. A DNA structural Comprehensive analysis of hydrogen bonds in
46. Boguski MS. Biosequence exegesis. Science
atlas for Escherichia coli. J Mol Biol 2000; 299 regulatory protein-DNA complexes: in search
1999; 286 (5439): 453-5.
(4): 907-30. of common principles. J Mol Biol 1995; 253
47. Miller C, Gurd J, Brass A. A RAPID
28. Kanehisa M; Goto S. KEGG: kyoto encyclo- algorithm for sequence database comparisons: (2): 370-82.
pedia of genes and genomes. Nucleic Acids application to the identification of vector 66. Luscombe NM, Laskowski RA, Thornton JM.
Res 2000; 28 (1): 27-30. contamination in the EMBL databases. Bio- Protein-DNA interactions: a 3D analysis of
29. Jeffery CJ. Moonlighting proteins. TIBS 1999; informatics 1999; 15 (2): 111-21. amino acid-base interactions. Nucleic Acids
24 (1): 8-11. 48. Gonnet GH, Korostensky C, Brenner S. Res. In Press.
30. Chothia, C. Proteins. One thousand families Evaluation measures of multiple sequence 67. Mandel-Gutfreund Y, Margalit H. Quantita-
for the molecular biologist. Nature 1992; 357 alignments. J Comput Biol 2000; 7 (1-2): tive parameters for amino acid-base inter-
(6379): 543-4. 261-76. action: inplications for prediction of protein-
49. Orengo CA, Taylor WR. SSAP: sequential DNA binding sites. Nucleic Acids Res 1998;
31. Orengo CA, Jones DT, Thornton JM. Protein
structure alignment program for protein struc- 26: 2306-12.
superfamilies and domain superfolds. Nature
1994; 372 (6507): 631-4. ture comparison. Methods Enzymol 1996; 266: 68. Sternberg MJ, Gabb HA, Jackson RM. Predic-
617-35. tive docking of protein-protein and protein-
32. Lesk AM, Chothia C. How different amino DNA complexes. Curr Opin Struct Biol 1998;
acid sequences determine similar protein 50. Orengo CA. CORA – topological fingerprints
for protein structural families. Protein Sci 8 (2): 250-6.
structures: the structure and evolutionary 69. Aloy P, et al. Modelling repressor proteins
dynamics of the globins. J Mol Biol 1980; 136 1999; 8 (4): 699-715.
51. Russell RB, Sternberg MJ. Structure predic- docking to DNA. Proteins 1998; 33 (4):
(3): 225-70. 535-49.
tion. How good are we? Curr Biol 1995; 5 (5):
33. Russell RB, et al. Recognition of analogous 488-90. 70. Dickerson RE. DNA-binding: the prevalence
and homologous protein folds: analysis of 52. Martin AC, et al. Protein folds and functions. of kinkiness and the virtues of normality.
sequence and structure conservation. J Mol Structure 1998; 6 (7): 875-84. Nucleic Acids Res 1998; 26 (8): 1906-26.
Biol 1997; 269 (3): 423-39. 53. Hegyi H, Gerstein M. The relationship be- 71. Perez-Rueda E, Collado-Vides J. The reper-
34. Russell RB, et al. Recognition of analogous tween protein structure and function: a com- toire of DNA-binding transcriptional regula-
and homologous protein folds – assessment of prehensive survey with application to the tors in Escherichia coli K-12. Nucleic Acids
prediction success and associated alignment yeast genome. J Mol Biol 1999; 288 (1): 147-64. Res 2000; 28 (8): 1838-47.
accuracy using empirical substitution matri- 54. Russell RB, Sasieni PD, Sternberg MJE. 72. Mewes HW, et al. MIPS: a database for geno-
ces. Protein Eng 1998; 11 (1): 1-9. Supersites within superfolds. Binding site mes and protein sequences. Nucleic Acids Res
35. Fitch WM. Distinguishing homologous from similarity in the absence of homology. J Mol 2000; 28 (1): 37-40.
analogous proteins. Syst Zool 1970; 19: 99-110. Biol 1998; 282 (4): 903-18. 73. Salgado H, et al. RegulonDB (version 3.0):
36. Tatusov RL, Koonin EV, Lipman DJ. A geno- 55. Wilson CA, Kreychman J, Gerstein M. As- transcriptional regulation and operon orga-
mic perspective on protein families. Science sessing annotation transfer for genomics: nization in Escherichia coli K-12. Nucleic
1997; 278 (5338): 631-7. quantifying the relations between protein Acids Res 2000; 28 (1): 65-7.
Method Inform Med 4/2001You can also read
