蛋白質(zhì)結(jié)構(gòu)預(yù)測-吳凌云

蛋白質(zhì)結(jié)構(gòu)預(yù)測-吳凌云.ppt目錄蛋白質(zhì)的三維結(jié)構(gòu)蛋白質(zhì)結(jié)構(gòu)預(yù)測的重要性蛋白質(zhì)二級結(jié)構(gòu)預(yù)測方法蛋白質(zhì)三級結(jié)構(gòu)預(yù)測方法研究趨勢蛋白質(zhì)的結(jié)構(gòu)sequencestructurefunctionmedicine蛋白質(zhì)三維結(jié)構(gòu)蛋白質(zhì)的三維結(jié)構(gòu)Alpha-helixBeta-sheetLoopandTurnTurnorcoil2007-5-18CUHK5蛋白質(zhì)的結(jié)構(gòu)層次蛋白質(zhì)二級結(jié)構(gòu)a-helix(30-35%)

a-螺旋b-sheet/b-strand(20-25%)

b-折疊Coil(40-50%)無規(guī)則卷曲Loop環(huán)b-turnb-轉(zhuǎn)角蛋白質(zhì)結(jié)構(gòu)分類法Class(<10)結(jié)構(gòu)類Folds(<1000)折疊子Superfamily超家族序列或結(jié)構(gòu)相似Family家族序列相似性>25%–30%同源Homology蛋白質(zhì)結(jié)構(gòu)類目錄蛋白質(zhì)的三維結(jié)構(gòu)蛋白質(zhì)結(jié)構(gòu)預(yù)測的重要性蛋白質(zhì)二級結(jié)構(gòu)預(yù)測方法蛋白質(zhì)三級結(jié)構(gòu)預(yù)測方法研究趨勢蛋白質(zhì)結(jié)構(gòu)與功能Hemoglobin(血紅蛋白)運(yùn)輸養(yǎng)氣蛋白質(zhì)結(jié)構(gòu)與功能Porin(孔蛋白)穿透細(xì)胞膜的運(yùn)輸?shù)鞍踪|(zhì)三維結(jié)構(gòu)的測定X-射線衍射法核磁共振法費(fèi)用昂貴速度慢對某些蛋白質(zhì)無法使用蛋白質(zhì)結(jié)構(gòu)預(yù)測從蛋白質(zhì)一級序列（氨基酸）序列預(yù)測蛋白質(zhì)三維結(jié)構(gòu)序列?二級結(jié)構(gòu)?三維結(jié)構(gòu)?功能必要性與可行性必要性DNA序列數(shù)據(jù)>>蛋白質(zhì)序列數(shù)據(jù)>>蛋白質(zhì)結(jié)構(gòu)數(shù)據(jù)意義重大：蛋白質(zhì)功能預(yù)測、藥物設(shè)計(jì)…可行性蛋白質(zhì)的序列信息唯一地決定三維結(jié)構(gòu)序列相似性意味著結(jié)構(gòu)相似性19941997200220052010序列(Swiss-Prot)40,00068,000114,033192,799515,203結(jié)構(gòu)(PDB)4,0457,00018,83832,43464,500序列—結(jié)構(gòu)—功能左邊的Bacteriocin是CASP-4測試中的T0102試題。中間的結(jié)果是Baker研究小組用從頭預(yù)測方法Rosetta獲得的。右邊是用中間的結(jié)構(gòu)在數(shù)據(jù)庫中進(jìn)行搜索得到的最相似的蛋白質(zhì)結(jié)構(gòu)。目錄蛋白質(zhì)的三維結(jié)構(gòu)蛋白質(zhì)結(jié)構(gòu)預(yù)測的重要性蛋白質(zhì)二級結(jié)構(gòu)預(yù)測方法蛋白質(zhì)三級結(jié)構(gòu)預(yù)測方法研究趨勢二級結(jié)構(gòu)預(yù)測Givenaproteinsequencea1a2…aN,secondarystructurepredictionaimsatdefiningthestateofeachaminoacidaiasbeingeitherH(helix),E(extended=strand),orO(other)(Somemethodshave4states:H,E,Tforturns,andOforother).Thequalityofsecondarystructurepredictionismeasuredwitha“3-stateaccuracy”score,orQ3.Q3isthepercentofresiduesthatmatch“reality”(X-raystructure).二級結(jié)構(gòu)的確定DetermineSecondaryStructurepositionsinknownproteinstructuresusingDSSPorSTRIDE:KabschandSander.DictionaryofSecondaryStructureinProteins:patternrecognitionofhydrogen-bondedandgeometricalfeatures.Biopolymer22:2571-2637(1983)(DSSP)FrischmanandArgos.Knowledge-basedsecondarystructureassignments.Proteins,23:566-571(1995)(STRIDE)Q3ALHEASGPSVILFGSDVTVPPASNAEQAKhhhhhooooeeeeoooeeeooooohhhhhohhhooooeeeeoooooeeeooohhhhhhhhhhhoooohhhhooohhhooooohhhhhAminoacidsequenceActualSecondaryStructureQ3=22/29=76%Q3=22/29=76%(usefulprediction)(terribleprediction)Q3forrandompredictionis33%Secondarystructureassignmentinrealproteinsisuncertaintoabout10%;Therefore,a“perfect”predictionwouldhaveQ3=90%.早期的二級結(jié)構(gòu)預(yù)測方法ChouandFasmanChouandFasman.Predictionofproteinconformation.Biochemistry,13:211-245,1974GORGarnier,OsguthorpeandRobson.Analysisoftheaccuracyandimplicationsofsimplemethodsforpredictingthesecondarystructureofglobularproteins.J.Mol.Biol.,120:97-120,1978ChouandFasmanStartbycomputingaminoacidspropensitiestobelongtoagiventypeofsecondarystructure:Propensities>1meanthattheresiduetypeiislikelytobefoundinthecorrespondingsecondarystructuretype.AminoAcidsPropensityAminoAcid

-Helix

-Sheet Turn

Ala 1.29 0.90 0.78Cys 1.11 0.74 0.80Leu 1.30 1.02 0.59Met 1.47 0.97 0.39Glu 1.44 0.75 1.00Gln 1.27 0.80 0.97His 1.22 1.08 0.69Lys 1.23 0.77 0.96

Val 0.91 1.49 0.47Ile 0.97 1.45 0.51Phe 1.07 1.32 0.58Tyr 0.72 1.25 1.05Trp 0.99 1.14 0.75Thr 0.82 1.21 1.03

Gly 0.56 0.92 1.64Ser 0.82 0.95 1.33Asp 1.04 0.72 1.41Asn 0.90 0.76 1.23Pro 0.52 0.64 1.91

Arg 0.96 0.99 0.88

Favorsa-HelixFavorsb-strandFavorsturnChouandFasmanPredictinghelices:findnucleationsite:4outof6contiguousresidueswithP(a)>1extension:extendhelixinbothdirectionsuntilasetof4contiguousresidueshasanaverageP(a)<1(breaker)ifaverageP(a)overwholeregionis>1,itispredictedtobehelicalChouandFasmanPredictingstrands:findnucleationsite:3outof5contiguousresidueswithP(b)>1extension:extendstrandinbothdirectionsuntilasetof4contiguousresidueshasanaverageP(b)<1(breaker)ifaverageP(b)overwholeregionis>1,itispredictedtobeastrandChouandFasmanPosition-specificparametersforturn:Eachpositionhasdistinctaminoacidpreferences.Examples:Atposition2,Proishighlypreferred;TrpisdisfavoredAtposition3,Asp,AsnandGlyarepreferredAtposition4,Trp,GlyandCyspreferredf(i)f(i+1)f(i+2)f(i+3)ChouandFasmanPredictingturns:foreachtetrapeptidestartingatresiduei,compute:PTurn(averagepropensityoverall4residues)F=f(i)*f(i+1)*f(i+2)*f(i+3)ifPTurn>PaandPTurn>Pb

andPTurn>1andF>0.000075tetrapeptideisconsideredaturn./fasta_www/chofas.htmPosition-dependentpropensitiesforhelix,sheetorturniscalculatedforeachaminoacid.Foreachpositionjinthesequence,eightresiduesoneithersideareconsidered.Ahelixpropensitytablecontainsinformationaboutpropensityforresiduesat17positionswhentheconformationofresiduejishelical.Thehelixpropensitytableshave20x17entries.Buildsimilartablesforstrandsandturns.ThepredictedstateofAAjiscalculatedasthesumoftheposition-dependentpropensitiesofallresiduesaroundAAj.GORcanbeusedat:(GOR5,2005)j二級結(jié)構(gòu)預(yù)測方法的發(fā)展基于堿基的統(tǒng)計(jì)信息Chou-Fasman,GOR復(fù)雜統(tǒng)計(jì)信息GOR3,Qian-Sejnowski同源序列信息PHD,PREDATOR模型假設(shè)：每個神經(jīng)元都是一個多輸入單輸出的信息處理單元；神經(jīng)元輸入分興奮性輸入和抑制性輸入兩種類型；神經(jīng)元具有空間整合特性和閾值特性；神經(jīng)元輸入與輸出間有固定的時滯,主要取決于突觸延擱；忽略時間整合作用和不應(yīng)期；神經(jīng)元本身是非時變的，即其突觸時延和突觸強(qiáng)度均為常數(shù)。神經(jīng)元的數(shù)學(xué)模型神經(jīng)元函數(shù)前向神經(jīng)網(wǎng)絡(luò)（BP網(wǎng)）回饋神經(jīng)網(wǎng)絡(luò)（Hopfield網(wǎng)）自組織神經(jīng)網(wǎng)絡(luò)（SOM）神經(jīng)網(wǎng)絡(luò)的分類BP神經(jīng)網(wǎng)絡(luò)PHD:輸入13x20=260valuesForeachresidue,considerawindowofsize13:PHD:Network1（序列結(jié)構(gòu)）13x20values3valuesPa(i)Pb(i)Pc(i)Network1PHD:Network2（結(jié)構(gòu)

結(jié)構(gòu)）3valuesPa(i)Pb(i)Pc(i)3valuesPa(i)Pb(i)Pc(i)17x3=51valuesForeachresidue,considerawindowofsize17:Network2二級預(yù)測服務(wù)器Availableservers:ChouandFasman:/fasta_www/chofas.htmGOR:PHD:/predictprotein/JPRED:pbio.dundee.ac.uk/~www-jpred/PSIPRED:http://bioinf.cs.ucl.ac.uk/psipred/NNPREDICT:/~nomi/nnpredict.html目錄蛋白質(zhì)的三維結(jié)構(gòu)蛋白質(zhì)結(jié)構(gòu)預(yù)測的重要性蛋白質(zhì)二級結(jié)構(gòu)預(yù)測方法蛋白質(zhì)三級結(jié)構(gòu)預(yù)測方法研究趨勢ModelingproteinfoldingSimpleexactmodel+approximatealgorithmLatticeModelsTwentyaminoacidscanbedividedintotwoclasses: Hydrophobic/Non-polar(H)(疏水)

Hydrophilic/Polar(P)(親水)ThecontactsbetweenHpointsarefavorablehydrophobicaminoacidhydrophilicaminoacidCovalentbondH-HcontactGoal:maximizethenumberofH-HcontactsHPModelHPModelReducecomputationbylimitingdegreesoffreedomLimita-carbon(Ca)atomstopositionson2Dor3DlatticeProteinsequence?

representedaspaththroughlatticepointsEmphasisonforminghydrophobiccoreComplexityAcombinatorialoptimizationproblemNP-hardproblemLongrangeinteraction+globaloptimizationGAMCSA-----timeconsumedHowBadareNP-CompleteProblems?Length=20SOMApproachExistedSOMsolutionMotivatedbySOMforTSPIncorporationofHPInformationCompactlatticeNewSOMApproachMotivationConsiderabiglatticeMultiplemapofSOMFeasibilityofsolutionsEquivalenttoPCTSPProperlydefinethelatticedistanceTSPforce+H-HforceNewSOMApproachApproachsInitializationLearningsamplesetpartitionstrategyLearningsamplesetreductionstrategyLocalsearchprocedureNumerical

ResultsConstructedHPsequencesHPbenchmark(upto36aminoacids)UniqueOptimalFoldingProblemWhatproteinsinthetwodimensionalHPmodelhaveuniqueoptimal(minimumenergy)folding?(BrianHayes,1998)OswinAichholzerprovedthatinsquarelatticeThereareclosedchainsofmonomerswiththispropertyforallevenlengths.Thereareopenmonomerchainswiththispropertyforalllengthsdivisiblebyfour.SquareLatticeandTriangularLatticeOurResultsForanyn=18k(kisapositiveinteger),thereexistsann-node(openorclosed)chainwithatleast3^O(n)optimalfoldingsallwithisomorphiccontactgraphsofsizen/2.On2Dtriangularlattice,foranyintegern>19,thereexistbothclosedandopenchainsofnnodeswithuniqueoptimalfolding.ExamplesofOptimalFoldings三級結(jié)構(gòu)預(yù)測方法Comparative(homology)modeling(比較建模法)通過與已知結(jié)構(gòu)蛋白質(zhì)的序列比對來預(yù)測（序列-序列比對）適用于>50%的未知結(jié)構(gòu)蛋白質(zhì)Threading(foldrecognition)(折疊識別法)從已知的結(jié)構(gòu)類中找到最適合的（序列-結(jié)構(gòu)比對）適用于30%左右的未知結(jié)構(gòu)蛋白質(zhì)Abinitio/denovomethods(從頭預(yù)測法)不使用已知的結(jié)構(gòu)信息，僅僅利用序列信息來預(yù)測適用于全新結(jié)構(gòu)的蛋白質(zhì)RMSD蛋白質(zhì)的空間坐標(biāo)表示定義氨基酸（原子）對應(yīng)關(guān)系：計(jì)算RMSD：比較建模法原理序列相似性>25%的蛋白質(zhì)具有相似的結(jié)構(gòu)基本步驟將目標(biāo)序列和模板序列進(jìn)行比對（關(guān)鍵步驟）構(gòu)造主干（backbone）的三維模型用二級結(jié)構(gòu)預(yù)測和手工預(yù)測等方法來填補(bǔ)一些空隙構(gòu)造環(huán)（loops）和卷曲（coils）安裝側(cè)鏈搜索側(cè)鏈數(shù)據(jù)庫，使用分子動力學(xué)優(yōu)化/檢驗(yàn)?zāi)Ｐ完P(guān)鍵方法序列-序列比對折疊識別法原理蛋白質(zhì)具有很少的折疊類型（<1000）基本步驟將目標(biāo)蛋白質(zhì)序列與已知的折疊進(jìn)行比對將目標(biāo)序列“安裝”到選擇的模板結(jié)構(gòu)上對模型進(jìn)行優(yōu)化、調(diào)整檢驗(yàn)?zāi)Ｐ偷暮侠硇躁P(guān)鍵方法序列-結(jié)構(gòu)比對折疊識別法折疊識別法動態(tài)規(guī)劃人工神經(jīng)網(wǎng)絡(luò)分支定界法線性規(guī)劃MonteCarlo方法從頭預(yù)測法理論基礎(chǔ)蛋白質(zhì)的天然構(gòu)象是熱力學(xué)最穩(wěn)定構(gòu)象、也是能量最低構(gòu)象能量函數(shù)通過原子間作用力計(jì)算出的熱力學(xué)能量精確，但難以計(jì)算偽能量函數(shù)根據(jù)已知的三維結(jié)構(gòu)知識得到的勢能函數(shù)常見的結(jié)構(gòu)?

低能量不常見的結(jié)構(gòu)?高能量極罕見的結(jié)構(gòu)?極高能量原子間作用力蛋白質(zhì)的勢能蛋白質(zhì)的勢能能量函數(shù)基于能量最小化的從頭預(yù)測法基本步驟選擇蛋白質(zhì)能量函數(shù)模型選擇三維結(jié)構(gòu)的表示方法選擇三維結(jié)構(gòu)的評價函數(shù)選擇尋找最優(yōu)結(jié)構(gòu)的優(yōu)化方法關(guān)鍵問題大規(guī)模非線性規(guī)劃大量的局部極小點(diǎn)在計(jì)算時間和準(zhǔn)確度之間尋找平衡最優(yōu)化方法非線性規(guī)劃最速下降法牛頓法共軛梯度法全局優(yōu)化方法遺傳算法分解-結(jié)合法離散化方法MonteCarlo方法Metropolis算法基本步驟選擇一個初始結(jié)構(gòu)A隨機(jī)產(chǎn)生一個新的結(jié)構(gòu)B如果，接受B否則產(chǎn)生一個[0,1]之間的隨機(jī)數(shù)r如果，接受B，否則拒絕重復(fù)以上步驟溫度T的修改——模擬退火方法分子動力學(xué)（動力系統(tǒng)）方法對蛋白質(zhì)中原子間的作用力進(jìn)行建模用動力學(xué)方程跟蹤蛋白質(zhì)折疊時每個原子的位置求解這些方程的解析解是極其困難的用離散動力系統(tǒng)的方法來確定數(shù)值解問題模擬蛋白質(zhì)折疊過