Comparative and Functional Genomics Comp Funct Genom 2001; 2: 330–337. DOI: 10.1002/cfg.108 Feature Meeting Review: The Intelligent Systems in Bioinformatics Conference 2001 (ISMB2001) 21st–25th July, Tivoli Gardens, Copenhagen K. Cara Woodwark* Biomolecular Sciences, UMIST, Manchester, M60 1QD, UK *Correspondence to: K. Cara Woodwark, Biomolecular Sciences, UMIST, PO Box 88, Manchester, M60 1QD, UK. Keywords: bioinformatics; conference; microarray; promoter; promoter prediction; gene E-mail: [email protected] prediction; RNA This year’s ISMB conference, an annual event explained that, given the right circumstances, any organised by the International Society for Compu- protein will form fibrils similar to those found in tational Biology (http://www.iscb.org) was the big- BSE or Alzheimer’s. gest ever, with over 1400 delegates. The venue was Proteins seem to fold and unfold all the time, also a first, situated, as it was, in Copenhagen’s which can cause problems for structure determina- Tivoli Gardens funfair. However there was still time tion in protein crystallography and NMR, as for the many satellite meetings that flanked the proteins made under different circumstances often conference including the Bio Pathways conference, have different structures. In the densely packed the Bioinformatics Open Source Conference and the environment of the cell, folding and unfolding may Bio-Ontologies Conference. form part of a switch mechanism, or chaperones Soren Brunak and Anders Krogh opened the may be involved to help a protein to fold into a conference. They remarked that it was 30 years particular structure. since the Needleman-Wunsch algorithm was writ- Aggregations or amyloid structures are respon- ten, but that things have not changed much, from sible for many diseases e.g. Alzheimers, New that time, in that the basic biological ideas are still variant CJD (related to BSE), Type II diabetes etc. driving research. They also mentioned that there Apparently, by the age of 60 we will all develop had been 180 papers submitted to the conference all some sort of aggregate, but hopefully they should of which had to be refereed and graded before be disease free (asymptomatic). 16 diseases caused choosing the 38 speakers. by amyloid structures have now been identified (20 Only the keynote talks are covered in depth here if diseases such as Parkinsons are included). For as all the other talks are covered in a special example, two point mutations in Lysozyme allow it supplement to the journal Bioinformatics. http:// to form disease-causing fibrils, composed of many bioinformatics.oupjournals.org/ parallel beta sheets. The conference began with talks on Protein The major breakthrough came when one of Structure and Modelling. However, rather than Chris’ students was working on PI3 Kinase NMR, talks about protein structure prediction, these talks when he went for a long weekend (160 hours). On were based much more on the biology of how his return, the trace had disappeared almost to protein structure information can help our under- nothing. So they looked to see what had happened standing of the evolution and function of proteins. to the protein and found that it had formed fibrils. Chris Dobson (Cambridge University) opened the This was a complete surprise as it wasn’t a disease conference with an excellent talk on Protein Fold- causing protein and so was not expected to form ing, Molecular Evolution, and Human Disease.He fibrils. After examining the fibrils they discovered Copyright # 2001 John Wiley & Sons, Ltd. Meeting Review 331 that they were hollow pipes formed by 4 groups of 2 becoming massive balls of fibrils as if we were to beta sheets wrapped around each other in a helical live long enough that is how we would end up! formation. These might be useful as nanotubes! All of this was discovered because someone had a Chris believes that the ability to form fibrils is long weekend! a character of all proteins, for example, even myo- The rest of the section was an interesting mix of globin, in its less soluble form, produced amyloid different aspects of protein structure. Gordana Apic fibres. In fact all proteins they have tried have (MRC Laboratory of Molecular Medicine) gave us produced fibrils, given the correct circumstances, an Insight into Domain Combinations. Potentially and any polypeptide chain if not chaperoned, or there are 180 000 pairwise combinations of SCOP controlled, could form fibrils. domains, but only 1,000 of these are found in There appears to be an initial time limiting step, 20 000 multidomain proteins from 40 species. as, like crystallisation, fibril formation needs a Indeed, 60% of domains have only one known nucleation, or seeding, step. This explains the combination partner. The domain order is highly rapid onset of diseases such as BSE after the first conserved within protein families. Stephen Mo¨ller symptoms are noticed, as after the initial contam- (EBI) spoke about predicting not only G protein ination with ‘‘seed’’ proteins there is a slow coupled receptors, but also their specificity, using ‘‘incubation’’ period until enough plaques are ‘‘SPEXS’’ (http://ep.ebi.ac.uk/, http://www.ebi.ac.uk/ formed to cause symptoms. After these first signs ycroning/coupling.html). Gianluca Pollastri (UC the growth of the fibrils takes place rapidly, Irvine & Bologna) used bi-directional neural network especially as the intermediate form of the fibril is architectures and evolutionary information to predict the most ‘‘contagious’’. Initial aggregates rather interaction positions between proteins, as structure than the fibrils are the real seeds. At this stage they tends to be more conserved than sequence (http:// are toxic and may lead to apoptosis, thus ridding promoter.ics.uci.edu/BRNN-PRED/). Tobias Mu¨ller the body of a diseased cell, although this is not (Deutsches Krebsforschungszentrum) also combined always a good thing, as even more ‘‘seed’’ forming structure prediction with sequence searching, but this fibrils may be released, to be taken up by other time used transmembrane domain specific matrices in cells. order to facilitate the search for homologous trans- The reason that age seems to be a factor in many membrane proteins (http://www.dkfz.de/tbi/people/ of these amyloid diseases is that over the years there tmueller). Michael Lappe (EBI) used a combination is an increased risk that something will go wrong of structure, in the form of fold information, and with the folding of a protein, thus forming a protein interaction data to predict function, nucleation seed. However, some proteins are more although the method is still very much in develop- likely to aggregate than others, due to sequence and ment (http://www.ebi.ac.uk/ylappe/FoldPred). cellular circumstance. For example, some mutated Chris Sander (Whitehead Institute) gave the proteins aggregate faster than others from a single keynote talk for the Sequence Motifs, Alignments point mutation. If, for example, the mutation is in and Families Section. He introduced his talk on the C terminus, then it is more likely to cause Structural Genomics, by pointing out what was to protein aggregation, even though it may not have be the overwhelming take-home message of the much affect on the folding. The position and conference – that what we (Bioinformaticians as ‘‘biology’’ of a protein is also important, as for well as Biologists) are trying to do is answer example, in vitro, myoglobin will aggregate more Biological questions. He hoped that soon we quickly than prions, yet there is no myoglobin would be able to model the ‘e-cell’ and then the aggregate disease. Therefore, whether a protein ‘e-organ’; model the perturbation of the system forms aggregates also depends on its position and caused by drugs; decipher neurobiology, and com- interactions within the cell. Selection against aggre- bine it all into systems biology where not just the gation may have increased the occurrence of whole organism, but whole ecosystems could be chaperones, as protein mixes do not form fibrils. modelled. Heterozygotes, with two alleles of a protein, may According to Chris Sander, structural genomics is also be at an advantage as mixed fibrils are also less only a part of what is needed. Eventually he would likely, so even sexual reproduction plays a role. like to be able to see the structure of every Chris believes that the whole of evolution and biological molecule, from which we would be able biology is one huge strategy to keep organisms from to fully understand the function of that molecule. Copyright # 2001 John Wiley & Sons, Ltd. Comp Funct Genom 2001; 2: 330–337. 332 Meeting Review However, protein structure is difficult and time (Epigenomics AG) differentiated between acute consuming to obtain. Sander suggests that rather lymphoblastic leukaemia (ALL) and acute myeloid than obtaining the structure of every molecule, we leukaemia (AML) by looking at the methylation need only determine the structure of one molecule state of CpG dinucleotides in CpG islands, which per protein family, or maybe even superfamily, are responsible for expression regulation. Their using homology to predict the structure of the other algorithm enabled the most discriminant sites to family members. be chosen. Giulio Pavesi (University of Milan) Even at the level of 30% sequence identity, there explained the huge numbers involved in promoter are 4000 sequences for all the families of the model prediction. For example, if a pattern of m letters is organisms in Pfam. However, not all families are in all sequences studied and may have mutations at represented in Pfam, so the actual number may be any position, then there are 4m possibilities. While closer to 4000r4. Producing this number of promoter regions are very small, e.g. m=6, this is a representative structures will take as much coordi- tractable number, 4096 possibilities, but with a nation as sequencing the human genome and to signal length of 20, then it is in to the trillions ensure availability to everyone, something that Bill (109,951,1627,776 to be exact).