BIOINFORMIS or why biologists need computers INRODUION O SEQUENE NLYSIS dot plots, alignments, and similarity searches http://www.bioinformatics.uni-muenster.de/teaching/courses/bioinf/index.hbi Prof. Dr. Wojciech Makałowski Institute of Bioinformatics EVOLUIONRY BSIS OF SEQUENE NLYSES HISISNNESRLSEQUENE EVOLUIONRY BSIS OF SEQUENE NLYSES HISISNNESRLSEQUENE HISISNMNESRLSEQUENE 3 EVOLUIONRY BSIS OF SEQUENE NLYSES HISISNNESRLSEQUENE HISISNMNESRLSEQUENE HISISNMNESRWSEQUENE EVOLUIONRY BSIS OF SEQUENE NLYSES HISISNNESRLSEQUENE HISISNMNESRLSEQUENE HISISNMNESRWSEQUENE HISISNMPESRWSEQUENE bioinfo_3_ - November,
EVOLUIONRY BSIS OF SEQUENE NLYSES HISISNNESRLSEQUENE HISISNMNESRLSEQUENE HISISNMNESRWSEQUENE HISISNMPESRWSEQUENE HISISNMPESRWSEQUENE EVOLUIONRY BSIS OF SEQUENE NLYSES HISISNMPESRWSEQUENE ene duplication or speciation! HISISNMPESRWSEQUENE Please note deletion of 7 8 EVOLUIONRY BSIS OF SEQUENE NLYSES EVOLUIONRY BSIS OF SEQUENE NLYSES HISISNMPESRWSEQUENE HISISOMPEERWSEQUENE HISISOMPEELWSEQUENE HISISNMPESRWSEQUENE HISISNMPERSXRSEQUENE Please note deletion of and W compensated by insertion of R and X 9 HISISNMPEEXRSEQUENE Please note insertion of EVOLUIONRY BSIS OF SEQUENE NLYSES EVOLUIONRY BSIS OF SEQUENE NLYSES HISISOMPLELNWSEQUENE HISISOMPLELNWSEQUENE HISISSMPEEXRSEQUENE HISISSUPEEXRSEQUENE bioinfo_3_ - November,
EVOLUIONRY BSIS OF SEQUENE NLYSES HISISOMPLELNEWSEQUENE EVOLUIONRY BSIS OF SEQUENE NLYSES HISISOMPLEELYNEWSEQUENE HISISSUPEEXRSEQUENE HISISSUPEREXRSEQUENE Please note another deletion of and insertion of R 3 EVOLUIONRY BSIS OF SEQUENE NLYSES HUMN OLON NER ENE ND BERIL DN REPIR ENE 3Myr Myr Myr HISISOMPLEELYNEWSEQUENE HISISSUPEREXRSEQUENE Bacteria MSH_Human SPE_DROME MSH_Yeast MUS_EOLI Yeast Worm Fly MJOR EHNIQUES O BE DISUSSED Mouse Human VIVLMQIFVPESEVSIVDILRVDSQLKVSFMEMLESILRSK VVLMHIFVPSLISMVDSILRVSDNIIKLSFMVEMIESIIRD VVISLMQIFVPEEEIIVDILRVDSQLKVSFMVEILESILKNSK LILMYISYVPQKVEIPIDRIFRVDDLSRSFMVEMENILRNE *** ** ** * * **** **** * ** * * HOW O SOLVE HE PROBLEM HUMN OR OMPUER? very smart Dot Matrix plots not as cute looking as humans slow error prone doesn t like repetitive tasks Sequence alignments not so smart (stupid) extremely fast very accurate Similarity searches doesn t understand human languages; needs instruction provided in a special way 7 8 bioinfo_3_ - November,
LORIHM step-by-step problemsolving procedure, especially an established, recursive computational procedure for solving a problem in a finite number of steps. Dot Matrix Plots 9 DO MRIX PLOS DO MRIX PLOS Sensitive qualitative indicators of similarity Better than alignments in some ways rearrangements repeated sequences Rely on visual perception (not quantitative) Useful for RN structure Simplest method - put a dot wherever sequences are identical little better - use a scoring table, put a dot wherever the residues have better than a certain score (especially useful for amino acid sequence comparison) Or, put a dot wherever you get at least n matches in a row (identity matching, compare/word) Even better - filter the plot WINDOWED SORES LORIHM WINDOWED SORES EXMPLE. calculate a score within a window of a given size, for example six. plot a point if score is over a threshold (stringency), for example 7% 3. move the window over a given step, for example one. repeat step one to three till the end of sequence Let's compare two nucleotide sequences using following parameters:,, 3 bioinfo_3_ - November,
7 matching plot a dot 3 matching no action matching no action matching plot a dot 7 8 matching letter no action 7 matching plot a dot 9 3 bioinfo_3_ - November,
3 matching no action matching letter no action 3 3 matching plot a dot matching letter no action 33 3 matching plot a dot 3 matching no action 3 3 bioinfo_3_ - November,
matching letter no action matching no action 37 38 matching no action matching plot a dot 39 For similar sequences dots form a diagonal line bioinfo_3_ - November,
DO PLO - EXMPLES DO PLO - EXMPLES Rec DN sequence from Helicobacter pylori and Streptococcus mutant, window= match= Rec DN sequence from Helicobacter pylori and Streptococcus mutant, window= match= 3 DO PLO - EXMPLES DO PLO - EXMPLES Rec DN sequence from Helicobacter pylori and Streptococcus mutant, window= match= Rec DN sequence from Helicobacter pylori and Streptococcus mutant, window= match= DO PLO - EXMPLES DO PLO - EXMPLES Rec DN sequence from Helicobacter pylori and Streptococcus mutant, window=9 match= Rec DN sequence from Helicobacter pylori and Streptococcus mutant, window= match=8 7 8 bioinfo_3_ - November,
DO PLO - WH N YOU SEE HERE? Similar regions Repeated sequences DO PLO EXMPLES - REPES Repeated sequence in Escherichia coli ribosomal protein S Sequence rearrangements RN structures ene order 9 DO PLO EXMPLES - RERRNEMENS DO PLO EXMPLES - RN SRUURE deletion duplication omplementary region inversion Low complexity region DO PLO EXMPLES - ENE ORDER DO PLO EXMPLES - POENIL OPERONS Whole genome comparison of Buchnera against Wigglesworthia red dots - genes on the same strand green dots - genes on opposite strand Whole genome comparison of Buchnera against Wigglesworthia red dots - genes on the same strand green dots - genes on opposite strand 3 bioinfo_3_ - November,
DO PLO EXMPLES - PRLOOUS ENES Whole genome comparison of Wigglesworthia red dots - paralogs on the same strand green dots - paralogs on opposite strand Note: self-hits of all genes form red diagonal line DO PLOS RULES OF HUMB Don't get too many points, about 3 times the length of the sequence is about right (%) Window size about for distant proteins and for nucleic acid (try stringency %) heck sequence against itself Finds internal repeats heck sequence against another sequence Finds repeats and rearrangements he best programs should have dynamic adjustment of parameters dotlet: http://myhits.isb-sib.ch/cgi-bin/dotlet gepard: http://www.helmholtz-muenchen.de/icb/gepard DO PLOS VERSUS LINMENS LINMEN Linear representation of relation between sequences that shows one-to-one correspondence between amino acid or nucleotide residue How can we define a quantitative measure of sequence similarity? match mismatch gap gctg-aa-cg -ctataa-tc 7 8 LINMEN PROBLEM LINMEN PROBLEM HISISOMPLEELYNEWSEQUENE HISISSUPEREXRSEQUENE HISISNNESRLSEQUENE HISISOMPLEELYNEWSEQUENE HISISNNESRLSEQUENE HISISSUPEREXRSEQUENE 9 bioinfo_3_ - November,
LINMEN PROBLEM LINMEN PROBLEM HISISNNES-R--LSEQUENE HISISOMP-LEELYNEWSEQUENE HISISNNES-RLSEQUENE HISISSU-PEREXR-SEQUENE HISISOMP-LEELYNEWSEQUENE HISISNNES-R--LSEQUENE HISISNNES-RLSEQUENE HISISSU-PEREXR-SEQUENE LINMEN PROBLEM LINMEN PROBLEM HISISOMP-LE-ELYNEWSEQUENE HISISNNES--R--LSEQUENE HISISNNES--R--LSEQUENE HISISSU-PEREX-R---SEQUENE HISISOMP-LE-ELYNEWSEQUENE HISISSU-PEREX-R---SEQUENE he problem is that we need to model evolutionary events based on extant sequences, without knowing an ancestral sequence! 3 LINMEN ny assignment of correspondences that preserves the order of residues within the sequence is an alignment It is the basic tool of bioinformatics omputational challenge - introduction of insertions and deletions (gaps) that correspond to evolutionary events We must define criteria so that an algorithm can choose the best alignment LINMEN N EXMPLE Let's compare two strings gctgaacg and ctataatc an uninformative alignment -------gctgaacg ctataatc------- an alignment without gaps gctgaacg ctataatc an alignment with gaps gctga-a--cg --ct-ataatc another alignment with gaps gctg-aa-cg -ctataa-tc bioinfo_3_ - November,
SORIN SHEMES scoring system must account for residue substitution, and insertions or deletions (indels) Indels (gaps) will have scores that depend on their length For nucleic acid sequences, it is common to use a simple scheme for substitutions, e.g. for a match, for a mismatch More realistic would be to take into account nucleotide frequencies (sequence composition) and fact that transitions are more frequent than transversions 7 8 P SORIN SYSEMS P SORIN N EXMPLE non-affine gapping score - the second alignment is "better" non-affine model - each gap position treated the same, e.g. match =, mismatch =, gap affine model - first gap position penalized more than others, e.g. match =, mismatch =, gap opening = -8, gap = ------ = 3 ------ = 9 7 P SORIN N EXMPLE P SORIN N EXMPLE affine gapping score - the first alignment is "better" ------ = 7 Equivalent alignments ------ = ------ = ------ = 7 7 bioinfo_3_ - November,
MINO ID SORIN SYSEMS more complicated than nucleotide matrices first, we can align two homologous protein sequences and count the number of any particular substitution, for instance Serine to hreonine a likely change should score higher than a rare one we have to take into account that several the same position mutated several times after sequence divergence - this could bias statistics MINO ID SORIN SYSEMS to avoid this problem one can compare very similar sequences so one can assume that no position has changed more than once Margret Dayhoff introduced the PM system (Percent of ccepted Mutations) PM - two sequence have 99% identical residues PM - two sequence have 9% identical residues 73 7 PPROXIME RELION BEWEEN PM ND SEQUENE IDENIY PM MRIX LULION PM 3 8 sequence identity (%) 7 PM matrix is expressed as log-odds values multiplied by simply to avoid decimal points score of substitution i <-> j = log observed i <-> j mutation rate mutation rate expected from amino acids frequencies For instance, a value implies that in related sequences the mutation would be expected to occur. times more frequently than random. he calculation: he matrix entry corresponds to the actual value. because of the scaling. he value. is log of the relative expectation value of the mutation. herefore, the expectation value is. =. 7 7 MINO ID MRIES Problem with PM schema lies in that the high number matrices are extrapolated from closely related sequences Henikoffs developed the family of BLOSUM matrices based on the BLOKS database of aligned protein sequences, hence the name BLOcks SUbstitution Matrix observed substitution frequencies taken from conserved regions of proteins (blocks), not the whole proteins as in case of Dayhoff's work two avoid overweighting closely related sequences, the Hennikoffs replaced groups of proteins that have sequence identities higher than a threshold by either a single representative or a weighted average, e.g. for the commonly used BLOSUM matrix the threshold is % NOE reversed numbering of PM and BLOSUM matrices BLOSUM SORIN MRIX R N D 9 Q E H 8 I L K M F P 7 S some replacement are more frequent than others score system based on comparison of homologous domains W Y 3 7 V 3 R N D Q E H I L K M F P S W Y V 77 78 bioinfo_3_ - November,
BLOSUM SORIN MRIX R N D Q E H I L K M F P S W Y V R N D 9 Q E identical amino acids get positive score but not the same - frequent amino acids, e.g. alanine, get lower score than rare amino acids, e.g. tryptophan 8 H 3 I L K M 3 F 7 7 P S W Y V 79 R N D 9 Q substitutions to amino acids of different properties give a negative score E 8 H 3 I L K R N D Q E H I L K M F P S W Y V R N D 9 Q substitutions to amino acids of similar properties give a positive score E 8 H 3 I L K M 3 F 7 7 P S W Y V 8 BLOSUM SORIN MRIX R N D Q E H I L K M F P S W Y V BLOSUM SORIN MRIX SORIN REOMMENDIONS nucleotide sequence comparison match, mismatch, gap opening, gap extension amino acid sequence comparison M 3 F 7 7 P S W Y for general use (e.g. unknown sequence similarity) - BLOSUM V 8 8 LOBL VERSUS LOL LINMEN Optimal global alignment Optimal local alignment Sequences align essentially from end to end. Needleman & for diverged proteins - PM or BLOSUM3 for similar sequences - PM or BLOSUM8 Sequence alignment using dynamic programming Sequences align only in small, isolated regions. Smith & Waterman Wunsch (97) (98) 83 8 bioinfo_3_ - November,
onstruct an optimal alignment of these two sequences: Using these scoring rules: ap: rrange the sequence residues along a twodimensional lattice Vertices of the lattice fall between 8 8 he goal is to find the optimal path from here to here Each path corresponds to a unique alignment Which one is optimal? 87 88 he score for a path is the sum of its incremental edges scores ap: aligned with Match = he score for a path is the sum of its incremental edges scores ap: aligned with Mismatch = 89 9 bioinfo_3_ - November,
he score for a path is the sum of its incremental edges scores ap: aligned with NULL ap = NULL aligned with z Incrementally extend the path ap: 9 9 Incrementally extend the path Remember the best sub-path leading to each point on the lattice ap: Incrementally extend the path Remember the best sub-path leading to each point on the lattice ap: 93 9 Incrementally extend the path Remember the best sub-path leading to each point on the lattice ap: Incrementally extend the path Remember the best sub-path leading to each point on the lattice ap: 9 9 bioinfo_3_ - November,
Incrementally extend the path Remember the best sub-path leading to each point on the lattice ap: Incrementally extend the path Remember the best sub-path leading to each point on the lattice ap: - -7-8 - + - -7 + 97 98 Incrementally extend the path Remember the best sub-path leading to each point on the lattice ap: - -7-8 - + - -7 + Print out the alignment - 99 Incrementally extend the path Remember the best sub-path leading to each point on the lattice ap: - -7-8 - + - -7 + Incrementally extend the path Remember the best sub-path leading to each point on the lattice ap: - -7-8 - + - -7 + bioinfo_3_ - November,
Print out the alignment - - Both alignments are optimal - give the same max. score SEQUENE SIMILRIY SERH 3 BSIS OF DBSE SERH Database searching is fundamentally different from alignment he goal is to find homologous sequences (often more than one), not to establish the correct one-to-one mapping of particular residues Usually, this is a necessary first step to making an information map between two sequences Database searching programs were originally thought of as approximations to dynamic programming alignments ssumption: the best database search conditions are those that would produce the correct alignment Key idea - most sequences don t match. If one can find a fast way to eliminate sequences that don't match, the search will go much faster BSIS OF DBSE SERH basic terminology: query - sequence to be used for the database search subject - sequence found in the database that meets some similarity criteria hit - local alignment between query and subject Related sequences have "diagonals" with high similarity BLS Basic Local lignment Search ool References: ltschul, S.F., ish, W., Miller, W., Myers, E.W. & Lipman, D.J. (99) "Basic local alignment search tool." J. Mol. Biol. :3. ltschul, S.F., Madden,.L., Schäffer,.., Zhang, J., Zhang, Z., Miller, W. & Lipman, D.J. (997) "apped BLS and PSI-BLS: a new generation of protein database search programs." Nucleic cids Res. :3389 7 8 bioinfo_3_ - November,
NULEOIDE BLS LORIHM.Break down query sequence into overlapping words..scan databases for exact matches of size W (BLSn) or pattern (MegaBlast). 3.ry to extend the word matches into the complete maximal scoring pair (MSP). Significance is easily calculated from Karlin- ltschul equation..perform local dynamic programming alignment around MSP regions 9 BLS - Maximal Segment Pairs (MSP) Highest scoring pair of identical length segments from two sequences Local alignment without gaps Expected distribution is known! 373 :: :::::: :: : potential MSP running sum match = mism. = BLS - extend word matches Most expensive step in BLS algorithm Extend to end of high scoring segment pair, or HSP. HSPs approximate maximal segment pairs or MSPs. hey are only approximate because extension does not continue until running score reaches zero - drop off value concept. fter initial hit was found BLS tries so called extension - an alignment is extended until the maximum value of the score drops by x, hence name x dropoff value PROEIN BLS LORIHM Break down query sequence into overlapping words and create a lookaup table. For each word, determine a neighborhood of words that, if found in another sequence, would likely to be part of a significant maximum scoring pair (MSP). Scan databases for neighborhood words. If two words are found on the same diagonal within a specified distance, try to extend the word matches into the complete MSP. Significance is (relatively) easy calculated from Karlin-ltschul equation. Perform local dynamic programming alignment around MSP regions first step of BLSp is controlled by three parameters and a score matrix w - word length (k-tuple in FS terminology); default value is 3 (lowest possible is ); two words on the same diagonal are required f - score threshold; unlike FS BLS allows mismatches at this step but overall score of the "mini-alignment" has to be above the threshold - the concept of "neighborhood words" BLSp - neighborhood words BLSp - neighborhood words Example - IV triplet hreshold f = (default for BLSp) f= Pairs marked in blue would initiate an alignment extension 3 bioinfo_3_ - November,
BLS - FINL SEP Smith-Waterman algorithm (local dynamic programming), discussed before but limited to regions that include the HSPs KRLIN-LHUL SISIS High scores of local alignments between two random sequences follow Extreme Value Distribution Significance of alignment with gaps can be evaluated using K and λ estimated from alignments of random sequences with same gap penalty and scoring parameters In spite of claims of being mathematically rigorous these parameters can only be estimated empirically KRLIN-LHUL SISIS For ungapped alignments their expected number with score S or greater equals E = Kmne -λs K i λ, are parameters related to a search space and scoring system, and m, n represent a query and database length, respectively. Score can be transformed to a bit-score according to formula S = bitscore = (λs - lnk)/ln, then E = mn -S KRLIN-LHUL SISIS for ungapped alignments parameters K and λ are calculated algebraically but for gapped alignment a solid theory doesn't exist and these parameters are calculated by simulation which has to be run for every combination of scoring system including gap penalties therefore not all gap opening and extension score combinations are available more at http://www.ncbi.nlm.nih.gov/bls/ tutorial/ltschul.html 7 8 BLS - KNOWN PROBLEMS Significance is calculated versus theoretic distribution using Karlin-ltschul equation not real sequences. ssumes sequences are random ssume database is one long sequence length effects are not corrected for Statistics are very inaccurate for short queries (ca. characters). Be careful when you change BLS parameters, some of them should be coordinated, e.g. match/mismatch penalty and X- drop off value nucleotide BLS - default parameters tuned up for speed not sensitivity [otea, Veeramachaneni, and Makalowski (3) Mastering seeds for genomic size nucleotide BLS searches. Nucleic cids Res. 3(3):93] BLS LORIHM IMPLEMENON Program Query Database type blastn nt nt megablast nt nt blastp aa aa blastx nt aa tblastn aa nt tblastx nt aa blastseq nt, aa nt, aa 9 bioinfo_3_ - November,