版權(quán)說(shuō)明:本文檔由用戶提供并上傳,收益歸屬內(nèi)容提供方,若內(nèi)容存在侵權(quán),請(qǐng)進(jìn)行舉報(bào)或認(rèn)領(lǐng)
文檔簡(jiǎn)介
1、中的應(yīng)用鄧明華dengmh隱馬氏模型及其在生物信息學(xué)內(nèi)容提要什么是隱馬氏模型什么是隱馬模型隱馬氏模型理論生物信息學(xué)應(yīng)用舉例Multiple sequence alignment (MSA )基因FindinggGene expression clusteringCNV detection5/17/2011Bioinformatics math.pku2韋小寶的骰子(隱馬氏模型)兩種骰子,開(kāi)始以兩種骰子始以2/5的概率出千。的概率出千正常A:以以1/6/的概率出現(xiàn)每個(gè)點(diǎn)不正常B:5,6出現(xiàn)概率為3/10,
2、其它為1/10出千的隨機(jī)規(guī)律0.20.8AB5/17/2011Bioinformatics math.pku3韋小寶的骰子(隱馬氏模型)觀測(cè)到其一次投擲結(jié)果測(cè)到其次投擲結(jié)果問(wèn)題:請(qǐng)判斷韋小寶什么時(shí)候出千了?5/17/2011Bioinformatics math.pku4隱馬氏模型理論識(shí)別問(wèn)題識(shí)別問(wèn)題已知若干個(gè)隱馬氏模型及其參知若個(gè)隱馬模型其參數(shù)數(shù),對(duì)一個(gè)觀測(cè)樣本,決定它來(lái)自哪一個(gè)對(duì)個(gè)觀測(cè)樣本決定它來(lái)自哪個(gè)模型(如例子中的識(shí)別問(wèn)題)。解碼問(wèn)題由觀測(cè)樣本得到隱狀態(tài);解碼問(wèn)題由觀測(cè)樣本得到隱狀態(tài)學(xué)習(xí)問(wèn)題由觀測(cè)樣本得到參數(shù)組 ;5/17/2011Bioin
3、formatics math.pku5隱馬氏模型的數(shù)學(xué)模型隱過(guò)程為X=X1,L,XT 觀察過(guò)程為Y=Y1,L,YYT模型參數(shù) = , A,B 初始分布=(i) , i=PX1=i轉(zhuǎn)移矩陣A= (aij ), aij = P(X(n+1=jj | Xn= i)給定某個(gè)時(shí)間的隱狀態(tài)的條件下, 觀測(cè)的分布矩陣B=(bill ) , bill = P(Yn=l
4、0; | Xn= i) 。5/17/2011Bioinformatics math.pku6識(shí)別問(wèn)題在已知若干個(gè)模型及其參數(shù)的情況下,識(shí)別問(wèn)題就是一個(gè)對(duì)于給定樣本進(jìn)行Bayesian 判決的問(wèn)題。判決步驟: 根據(jù)參數(shù)求出在每一個(gè)模型中, 出現(xiàn)給定樣本的概率P(Y | ),) 歸一化就得到給定樣本來(lái)自每歸化就得到給定樣本來(lái)自每個(gè)模型的概率P ( | Y) 。利用Bayesian 原理,就可以得到最好模型猜測(cè)。5/1
5、7/2011Bioinformatics math.pku7觀測(cè)序列的概率計(jì)算枚舉復(fù)雜度2TN2T多項(xiàng)式復(fù)雜度算法多項(xiàng)式復(fù)雜度算法:前傳算法和后傳算法前傳算法和后傳算法5/17/2011Bioinformatics math.pku8前傳概率5/17/2011Bioinformatics math.pku9前傳算法(Forward Algorithm)初始化迭代結(jié)果5/17/2011Bioinformatics math.pku10后傳概率5/17/2011Bioinformatics
6、 math.pku11后傳算法(Backward Algorithm)初始化迭代結(jié)果5/17/2011Bioinformatics math.pku12解碼問(wèn)題(I)要解決的問(wèn)題:給定觀測(cè)序列Y =(y1,y2,L,yT), 如何給出隱狀態(tài)序列X0=(x01,x02,L,x0T) 單點(diǎn)最優(yōu)路徑最優(yōu)指:對(duì)任意的路徑最優(yōu)指對(duì)任意的X=(xX(1,x2,L,xT) 有5/17/2011Bioinformatics math.pku13解碼問(wèn)題(II)由Bayesian公式有又由于序列Y
7、給定, 問(wèn)題等價(jià)于找最優(yōu)的X0使聯(lián)合概率Pr(x1,xx2,L,xxT;Ty1,yy2,L,yyT)最大。5/17/2011Bioinformatics math.pku14最優(yōu)單點(diǎn)確定5/17/2011Bioinformatics math.pku15Viterbi算法(I)算法的思想動(dòng)態(tài)規(guī)劃的遞推算法。遞推變量為我們有遞推公式以t(i)記錄t時(shí)刻時(shí)使t(j)aji最大的狀態(tài)j。5/17/2011Bioinformatics math.pku16Viterbi算法(II)初始化迭代5/17/2011Bioinform
8、atics math.pku17Viterbi算法(III)終止終后推5/17/2011Bioinformatics math.pku18Viterbi算法實(shí)例(I)轉(zhuǎn)移概率以及初概率轉(zhuǎn)移概率以初概率ABA0.80.2B0.10.9初概率0.60.4條件概率(Emission Probability)V1V2V3V4V5V6A1/61/61/61/61/61/6B5/17/2011Bioinformatics math.pku19Viterbi算法實(shí)例(II)ytt(A)t(A)t(
9、B)t(B)t=111.000x10-1-4.000x10-2-t=231.333x10-2A3.600x10-3Bt=341.778x10-3A3.240x10-4Bt=453.370x10-4A1.067x10-4At=553.161x10-4A2.880x10-5Bt=664.214x10-66A7.776x10-66Bt=765.619x10-7A2.100x10-6Btt=8837.492x107492x10-88A1.890x101890x10-77Bt=929.989x10-9A1.701x10-8Bt=1061.322x101322x10-9A4.592x104592x10-9
10、B5/17/2011Bioinformatics math.pku20Viterbi算法實(shí)例(III)觀測(cè)序列為:11 3 4 5 5 6 6 3 2 6345566326AAAAAAAAAABBBBBBBBBB解碼出來(lái)的狀態(tài)序列為:A A A B B B B B B B5/17/2011Bioinformatics math.pku21HMM學(xué)習(xí)問(wèn)題學(xué)習(xí)問(wèn)題:就是由觀測(cè)估計(jì)模型參數(shù).學(xué)習(xí)的兩種情況:觀測(cè)鏈相應(yīng)的狀態(tài)鏈已知;觀測(cè)鏈相應(yīng)的狀態(tài)鏈未知。5/17/2011Bioinformatics math.pku22學(xué)習(xí)原則極大似
11、然估計(jì)(MLE)狀態(tài)鏈已知時(shí)狀態(tài)鏈未知時(shí)5/17/2011Bioinformatics math.pku23狀態(tài)鏈已知時(shí)的MLE5/17/2011Bioinformatics math.pku24簡(jiǎn)單優(yōu)化問(wèn)題5/17/2011Bioinformatics math.pku25參數(shù)估計(jì)(狀態(tài)已知)把從狀態(tài)i到下一個(gè)時(shí)刻轉(zhuǎn)移為狀態(tài)j的頻數(shù)記為Aij,可估計(jì)aij為同樣記從狀態(tài)i到同一時(shí)刻的觀測(cè)轉(zhuǎn)移的頻數(shù)為Bil,則可估計(jì)bil為5/17/2011Bioinformatics math.pku26參數(shù)估計(jì)評(píng)價(jià)隱M
12、arkov模型的狀態(tài)鏈要有充分長(zhǎng)的樣本(大數(shù)定律,以頻率代替概率)。不幸的是狀態(tài)鏈往往并不知道, 而只是可以得到估計(jì), 不修正地使用頻率估計(jì)會(huì)增加誤差, 且這種估計(jì)不穩(wěn)健。5/17/2011Bioinformatics math.pku27參數(shù)估計(jì)的EM思想當(dāng)狀態(tài)鏈未知時(shí),由于似然函數(shù)的計(jì)算中包含了對(duì)所有可能的狀態(tài)鏈的求和計(jì)算過(guò)大在實(shí)際中是所有可能的狀態(tài)鏈的求和,計(jì)算過(guò)大,在實(shí)際中是不可能被采用的為此,人們采取折衷的方案,構(gòu)造一個(gè)遞推算法造個(gè)遞推算法,使之能相當(dāng)合理地給出模型參數(shù)使之能相當(dāng)合理地給出模型參數(shù)的粗略估計(jì)其核心思想是:在當(dāng)前參數(shù)下
13、,用期望值當(dāng)成頻數(shù)“數(shù)數(shù)”,并用“頻率”估計(jì)概率。這實(shí)際上是一種EM迭代算法思想。5/17/2011Bioinformatics math.pku28EM算法實(shí)際上是E (期望) 與M (最大化)兩個(gè)步驟合起來(lái)構(gòu)成的算法,稱為EM算法EM算法是針對(duì)測(cè)量數(shù)據(jù)不完全時(shí), 求參數(shù)的一種近似于最大似然估計(jì)的統(tǒng)計(jì)方法。HMM 模型參數(shù)的估計(jì), 是EM算法的算法的一個(gè)個(gè)最常見(jiàn)且極有用的一種典型例子5/17/2011Bioinformatics math.pku29EM算法基本框架觀測(cè)數(shù)據(jù)Y缺失數(shù)據(jù)X完全
14、數(shù)據(jù)Z=(Y, X).EStep (取期望).Mstep (取極大).5/17/2011Bioinformatics math.pku30期望頻數(shù)(狀態(tài)未知)5/17/2011Bioinformatics math.pku31期望頻數(shù)(狀態(tài)未知)5/17/2011Bioinformatics math.pku325/17/2011BaumWelch公式Bioinformatics math.pku33BaumWelch公式的推導(dǎo)(1)我們定義一個(gè)描述模型“趨勢(shì)”的量,以衡量參數(shù)
15、估計(jì)前后的概率分布的差異衡量參數(shù)估計(jì)前后的概率分布的差異。相對(duì)熵是最好的選擇5/17/2011Bioinformatics math.pku34相BaumWelch公式的推導(dǎo)(2)說(shuō)明只要依據(jù)Q函數(shù)增大來(lái)更新參數(shù),就能夠使得似然函數(shù)朝變大的方向改進(jìn),而且一個(gè)觀測(cè)Y對(duì)應(yīng)了一次改進(jìn)。5/17/2011Bioinformatics math.pku35BaumWelch公式的推導(dǎo)(3)于是要想得到參數(shù)修改的遞推公式, 只要把模型m修改為更好的模型m+1, 即只需將它取得使下式成立,5/17/2011Bioinformatics
16、; math.pku36BaumWelch公式的推導(dǎo)(4)寫(xiě)出出Q函數(shù)的表達(dá)式,我們有數(shù)的表達(dá)式我們有5/17/2011Bioinformatics math.pku37BaumWelch公式的推導(dǎo)(5)對(duì)每個(gè)變量可以分別取最大值。5/17/2011Bioinformatics math.pku385/17/2011初概率的重估計(jì)Bioinformatics math.pku39轉(zhuǎn)移概率重估計(jì)5/17/2011Bioinformatics math.pku40觀測(cè)概率重估計(jì)5/17/2011Bio
17、informatics math.pku41幾點(diǎn)說(shuō)明更詳細(xì)的內(nèi)容可參見(jiàn)應(yīng)用隨機(jī)過(guò)程第十章HMM.HMM在上面所述的算法中,初始值0 的設(shè)置會(huì)直接影響到估計(jì)的好壞為此常用的直接影響到估計(jì)的好壞為此,常用的一種方法是, 根據(jù)先驗(yàn)知識(shí)設(shè)置一條較長(zhǎng)的“標(biāo)準(zhǔn)虛擬”狀態(tài)鏈, 再用前面講的已知觀測(cè)鏈相應(yīng)的狀態(tài)鏈的情況下參數(shù)估計(jì)的方法, 得到一個(gè)對(duì)參數(shù)的粗估計(jì),并以它作為0的取值。的取值5/17/2011Bioinformatics math.pku42隱馬氏模型的生物信息學(xué)應(yīng)用Multiple sequence a
18、lignment基因FindingGene expression clusteringCNV detection5/17/2011Bioinformatics math.pku )43 (MSA應(yīng)用I: Multiple Sequence AlignmentKrogh A. et al, Hidden Markov model in computational biology. Applicat
19、ions to protein modeling. Journal of Molecular Biology 235:15011531, 1994.Eddy SR. Hidden Markov models. Current Opinion Structure Biology 6:361365, 1996.Eddy SR. Profile hidden M
20、arkov models. Bioinformatis14:755763, 1998.This part is modified from Colin Cherrys slides download fromwebdocswebdocs.cs.ualberta.ca/csualbertaca/colinc/cmput606/HMMOct25.Oct25ppt5/17/2011Bioinformatics math.pku 44inM
21、ethods for Characterizing a Protein FamilyObjective:sequences,bji Givencommonother members in i a numberb off relatedld encapsulatesuch of a waythe family. that what we they
22、; can have recognize in Some standard methods for characterization:Multiple AlignmentsRegular ExpressionsConsensus SequencesHidden Markov Models5/17/2011Bioinformatics math.pku45 A Characterization Exampl
23、e borrowed from SalzbergSalzberg, 19985/17/2011How(hypothetical) could wesequences? familycharacterize of nucleotide this Keep the Multiple AlignmentTry a regular expressionAT CG AC ACTG*&
24、#160;A TG GCBut what about?T G C T A G G vrsA C A C A T CTry a consensus sequence:A C A A T CDepends on distance measure math.pku46ExampleBioinformat
25、icsHMMs ProbabilitiesTransition probabilities5/17/2011Bioinformatics math.pku47EmissionInsert (Loop) States 5/17/2011 Bioinformatics math.pku 48 Scoring our simple HMM#1 “T G C T A G G” vrs:#2 “A
26、60;C A C A T C”Regular Expression (AT CG AC ACTG* A TG GC): #1 = Member #2: MemberHMM: #1 = Score of 0.0023% #2 Score of 4.7% (Probability)#1 =
27、 ScoreS off 0.97097#2 SScore off 66.77 (L(Log odds)dd)5/17/2011Bioinformatics math.pku49Standard Profile HMM ArchitectureThreeTh types off states:MatchInsertDeleteOne delete and one match per
28、60;position in modelOne insert per transition in modelStart and end “dummy” statesExample borrowed from Cline, 19995/17/2011Bioinformatics math.pku50Match StatesExample borrowed from Cline,
29、;19995/17/2011Bioinformatics math.pku51Insert StatesExample borrowed from Cline, 19995/17/2011Bioinformatics math.pku52Delete States Example borrowed from Cline, 1999 5/17/2011 Bioinformatics math.pku 53 Aligning
30、;and Training HMMsTraining from a Multiple AlignmentAligning a sequence to a modelCan be used to create an alignmentCan be used to score a sequenceqCan be used to in
31、terpret a sequenceTraining from unaligned sequences5/17/2011Bioinformatics math.pku54Training from an existing alignmentThis process what weve been seeing up to this point.Start with a
32、;predetermined number of states in your HMM.For each position in the model, assign a column in the multiple alignment that is relatively conserved.Emission probabilities are s
33、et according to amino acid counts in columns.Transition pprobabilities are set accordingg to how manyy sequences make use of a given delete or insert state.5/17/2011Bioinformatics&
34、#160; math.pku55Remember the simple exampleChose six positions in model.Highlighted area was selected to be modeled by an insert due to variability.yCan also do neat tricks fo
35、r picking length of model, such as model pruning.5/17/2011Bioinformatics math.pku56應(yīng)用II:Gene FindingBurge, C. and Karlin, S. (1997) Prediction of p ggene structures in human ggeno
36、mic completeDNA. J. Mol. Biol. 268, 7894Burge,Burge CC. BB. and Karlin,Karlin S.S (1998) Finding the genes in genomic DNA. Curr. Opin. Struct. Biol.8,346354. This part is
37、;modified from slides download fromwww.cs.ubc.ca/rogic/GeneFinding.ppt5/17/2011Bioinformatics math.pku57應(yīng)用II: Gene Finding5/17/2011Bioinformatics math.pku58Signals:g PremRNA SplicingpgStartcodonStart codonStopcodonStop codonGenomic
38、DNApre-mRNACap-Transcription-Poly(A)SplicingmRNAPtiProteinexonintronGTGT AGAGDonor siteAcceptorsiteCap-Poly(A)TranslationSplice sites5/17/2011Bioinformatics math.pku59Spliced AlignmentCompare withh cDNAor EST probesbGenomic DNAmRNA5/17/2011Start codonStop codonCap
39、-Poly(A)Poly(A)5-UTR3-UTRBioinformatics math.pku60Spliced AlignmentCCompare withith proteinti probesbGenomicGenomic DNADNAProtein5/17/2011Start codonStop codonBioinformatics math.pku61A eukaryotic geneThis is the human p53
40、tumor suppressor gene on chromosome 17.GenscanGi one ofisf ththe mostt popularl gene prediction algorithms.5/17/2011Bioinformatics math.pku62A eukaryotic geneIntronsFinal exon33 untranslated Initial&
41、#160;exonregionInternal exonsThis particular gene lies on the reverse strand.5/17/2011Bioinformatics math.pku63An Intronrevcomp(CT)=AGGT: signals startof intronAG: signals endof intronrevcomp(AC)=GT3 spliceli&
42、#160;siteit5 splice site5/17/2011Bioinformatics math.pku64Signals vs contentsIn gene finding, a small pattern within the genomicDNA is referred to as a signal, whereas a region of
43、 genomic DNA is a content.Examples of signals: splice sites, starts and ends of transcription or translation, branch points, transcription factor binding sitesExamples of contents:
44、 exons, introns, UTRs, promoterp regionsg5/17/2011Bioinformatics math.pku65 Prior knowledgeThThemultiple translated ofl 3.d regioni must haveh a lengthlh thath isi a SomeS codonsd are&
45、#160;more common thanh others.hExons are usually shorter than introns.The translated region begins with a start signal and ends with a stop codon.5 splice sites (exon to
46、intron) are usually GT; 3 splice sites (intron to exon) are usually AG.The distribution of nucleotides and dinucleotides isusually different in introns and exons.5/17/2011Bioinformatics
47、 math.pku66 Prior knowledgeWe want to build a probabilistic modelof a gene p our pprior knowledgeg.that incorporatesE.g., the translated region must have a length that i
48、s a multiple of 3.35/17/2011Bioinformatics math.pku67Prokaryotic Vs. Eukaryotic Gene FindingProkaryotes:small genomes 0.5 10106bphigh coding density (>90%)no intronsEukaryotes:large genomes 1071010bplo
49、w coding density (<50%)intron/exonstructureGene identification relatively easy, Gene identification a complex problem, with success rate 99%gene level accuracy 50%Problems:overlapping ORFsshort
50、60;genesfinding TSS and promoters5/17/2011Problems:manyBioinformatics math.pku68HMMs and Gene StructureNucleotides A,C,G,Tare the observablesDifferent states generates generate nucleotides at different frequen
51、ciesA simple HMM for unsplicedgenes:AAAGCATGCAT TTA ACG AGA GCA CAA GGG CTCTAATGCCGThe sequence of states is an annotation of the generated string each nucleotide is generated
52、 in intergenic, start/stop, codingstate5/17/2011Bioinformatics math.pku69Examples of Gene Finders Using HMMGGeneMarkMkHMMsHMM enhancedhd withih ribosomalibl bibindingdi sitei recognitioniiGenieneural networks&
53、#160;for splicing, HMMs for coding sensors, overall structure modeleddld bby HMMGenscanWM, WA and decision trees as signal sensors, HMMs for content sensors, overallll HMMHMMgeneHMM
54、0;trained using conditional maximum likelihoodMorgandecision trees for exonclassification, also Markov ModelsVEILsubHMMs each to describe a different bit of the sequence, overall HMM5/17/2011Bioi
55、nformatics math.pku70EXAMPLEEXAMPLE: fifindingdi genes withith VEILThe ViterbiExonIntronLocator(VEIL) was developed by JohnJohns Henderson, Hopkins University.University Steven Salzberg, and Ken Fas
56、manat Gene finder with a modular structure:Uses a HMM which is made up of subHMMs each to describenoncoding a differentDNA, exon, bit intron,of the sequence: upstr
57、eam Assumes test data starts and ends with noncodingDNA and contains exactly one gene.UsesUstart +biologicalbi stopl icodons,l kknowledge spliceld sites.tot “hardwire”“hdi” partt
58、;off HMM,HMM eg.5/17/2011Bioinformatics math.pku71 5/17/2011The exon submodelBioinformatics math.pku72Other submodelsThe start codonmodel is very simple:ExonThe splice junctions are also quite
59、simple and can behardwired (here is the 5 splice site):5/17/2011Bioinformatics math.pku73 The overall modelUpstreamStart codonEExonStop codon3 splice siteintron5 splice sitesiteFor more details, see J. Henderson, S.L. For more de
60、tails see J Henderson SL Salzberg, and K. Fasman(1997) Journal of Computational Biology 4:2, 127-141.5/17/2011Bioinformatics math.pku74Genscan ExampleDeveloped by Chris Burge 1997OneO off theth mostt accuratet abbinitioiitiprogramsUses
61、0;explicit state duration HMM to model gene structure (different length distributionsffor exons)Different model parameters for regions with different GC content5/17/2011Bioinformatics math.pku75
62、 E0I0Einit5 UTRE1I1EtermEsnglPNPE2I23 UTRpolyAforward strandbackward strandpolyA5 UTREsnglEinitI0E0I1E1EtermI2E23 UTR5/17/2011 math.pku76GenscansGenscans ArchitectureHMMs states ffor exons andd introns in threeh differentdff phases,h s
63、inglel exon, 5 and 3 UTRs, promoter region and polyA site and intergenic regionExplicitl lengthlh modelingdlHMMs for exons, introns and intergenic regionsWM and WA for accepto
64、r site, branch point, polyA site and promoter regionDecision tree (maximal dependence decomposition) for donor sites5/17/2011Bioinformatics math.pku77應(yīng)用III: Gene Expression ClusteringA. Schliepe
65、t al. Using hidden Markov model to analyze gene expression time course data. Bioinformatics 19(Suppl. 1) i255i263. 2003.M.Yuanand C.Kendziorski. Hidden Markov Models for Microarray
66、;Time Course Data in Multiple Biological Conditions (with Discussion)Discussion).Journal of the American Statistical Association101(476): 13231332; Discussion 13321340,1340 2006.(/talks/workshop
67、s/929103.2003/kendziorski/kendziorski_ima03.ppt) 5/17/2011Bioinformatics math.pku78 Alizadehet al., Nature 403:50311, 20005/17/2011Bioinformatics math.pku79連續(xù)觀測(cè)HMM5/17/2011Bioinformatics math.pku80Modelbased ClusteringiGiven n&
68、#160;sequences O, clustering is a partitionK clusters, each cluster is a hidden Markov modeldlJoint likelihood 5/17/2011Bioinformatics math.pku81Examples5/17/2011Bioinformatics math.pku82Iterative A
69、lgorithmIterationItti (t = 1,1 2,2 . . .)Generateassigning a new partitioning of the sequences by the likelihood each sequenceis Oi maximal.to the model k for whichCalcu
70、lateestimation newparameters algorithm parameters for
71、;using the reand each assigned model withsequence. their start Stop:function if thegrouping is belowimprovement a of the objective given iteration of the number sequen
72、cesgiven threshold, is reached does not change, the or a5/17/2011Bioinformatics math.pku83 應(yīng)用IV: CNV DetectionK. Wang et al. PennCNV: an integrated hidden g for
73、 highgresolution Markov model designedcopy number variation detection in wholegenome SNP genotyping datadata. Genome Research 17: 16651674, 2007.This part is modified from Yoon SooPyons
74、slides download from /images/6/6f/Cnv_snparray.ppt.5/17/2011Bioinformatics math.pku84What is CNV and LOHHomozygous deletionCopy number 0Copy neutral LOHCCopy numberb 25/17/2011deletionNormal num
75、ber 1Copy number 2amplicationCCopy numberb 6Bioinformatics math.pku85HemizygousCopyTwo methods of CNV identificationyClonehybridizationlbasedbd comparative (Arrayi CGH) genomici
76、;yTest and reference DNA are differentially fluorescent labeled and hybridized to the array.ycons: low resolution (Cannot find small CNV region)ySNP genotyping arrayypros: Higher resolu
77、tionyCons: poor signaltonoise ratio of hybridization5/17/2011Bioinformatics math.pku86Generation of SNP genotyping arrayIlumminaBead ArrayHuman1 Beadchip(100,000)240,000 BeadArray300,000550,000650,0001 Million just
78、;released. (human1M)AffymetrixSNP array10,000 (Mapping 10K array, 2003)100,000 (Mapping( 100K array)500,000 (Mapping 500K array)1 Million just released (Genomewide Human SNP 6.0)5/17/2011Bioinformatics math.pku87Genotyping Array5/17/2011Bioinformatics math.pku88SNP probeCAGACAGAAGTCTTGA/CAATCTATTTCTCATA.PMA: TGTCTTCAGAACTTAGATAAAGAGMMA: TGTCTTCAGAACTTAGATAAAGAGPMB:
溫馨提示
- 1. 本站所有資源如無(wú)特殊說(shuō)明,都需要本地電腦安裝OFFICE2007和PDF閱讀器。圖紙軟件為CAD,CAXA,PROE,UG,SolidWorks等.壓縮文件請(qǐng)下載最新的WinRAR軟件解壓。
- 2. 本站的文檔不包含任何第三方提供的附件圖紙等,如果需要附件,請(qǐng)聯(lián)系上傳者。文件的所有權(quán)益歸上傳用戶所有。
- 3. 本站RAR壓縮包中若帶圖紙,網(wǎng)頁(yè)內(nèi)容里面會(huì)有圖紙預(yù)覽,若沒(méi)有圖紙預(yù)覽就沒(méi)有圖紙。
- 4. 未經(jīng)權(quán)益所有人同意不得將文件中的內(nèi)容挪作商業(yè)或盈利用途。
- 5. 人人文庫(kù)網(wǎng)僅提供信息存儲(chǔ)空間,僅對(duì)用戶上傳內(nèi)容的表現(xiàn)方式做保護(hù)處理,對(duì)用戶上傳分享的文檔內(nèi)容本身不做任何修改或編輯,并不能對(duì)任何下載內(nèi)容負(fù)責(zé)。
- 6. 下載文件中如有侵權(quán)或不適當(dāng)內(nèi)容,請(qǐng)與我們聯(lián)系,我們立即糾正。
- 7. 本站不保證下載資源的準(zhǔn)確性、安全性和完整性, 同時(shí)也不承擔(dān)用戶因使用這些下載資源對(duì)自己和他人造成任何形式的傷害或損失。
最新文檔
- 養(yǎng)老院入住老人心理健康監(jiān)測(cè)制度
- 人力資源管理規(guī)范制度
- 企業(yè)內(nèi)部保密工作考核制度
- 后宮介紹教學(xué)課件
- 老年人護(hù)理知識(shí)分享守護(hù)長(zhǎng)者健康
- (人教版2026新教材)二年級(jí)英語(yǔ)下冊(cè)Unit 2單元測(cè)試(含解析+聽(tīng)力音頻)
- 皮革加工工安全宣教考核試卷含答案
- 熔體鎂工改進(jìn)模擬考核試卷含答案
- 尿素裝置操作工安全專項(xiàng)強(qiáng)化考核試卷含答案
- 我國(guó)上市公司獨(dú)立董事激勵(lì)制度的法律透視與完善路徑
- 光化學(xué)和光催化反應(yīng)的應(yīng)用
- 中日友好醫(yī)院公開(kāi)招聘工作人員3人筆試參考題庫(kù)(共500題)答案詳解版
- VDA6.3-2016過(guò)程審核主要證據(jù)清單
- 辦公耗材采購(gòu) 投標(biāo)方案(技術(shù)方案)
- 2020公務(wù)船技術(shù)規(guī)則
- 三片罐空罐檢驗(yàn)作業(yè)指導(dǎo)書(shū)
- 四川峨勝水泥集團(tuán)股份有限公司環(huán)保搬遷3000td熟料新型干法大壩水泥生產(chǎn)線環(huán)境影響評(píng)價(jià)報(bào)告書(shū)
- 開(kāi)庭陳述事實(shí)舉證范文(4篇)
- 管道焊接工藝和熱處理課件
- 2023深圳工務(wù)署品牌名單
- 二年級(jí)下冊(cè)課文快樂(lè)讀書(shū)吧-神筆馬良
評(píng)論
0/150
提交評(píng)論