29712.基于神經(jīng)網(wǎng)絡(luò)的蛋白質(zhì)二級(jí)結(jié)構(gòu)預(yù)測(cè)

1、北京科技大學(xué)碩士學(xué)位論文基于神經(jīng)網(wǎng)絡(luò)的蛋白質(zhì)二級(jí)結(jié)構(gòu)預(yù)測(cè)姓名：辛之歌申請(qǐng)學(xué)位級(jí)別：碩士專業(yè)：計(jì)算機(jī)科學(xué)與技術(shù)指導(dǎo)教師：楊炳儒20091110北京科技人學(xué)碩十學(xué)位論文PoretinSecodaryStructurePerdietionbasedonNeuralNetworkAbstractProteinISamajorbeareroflifeactivityanditsbiologicalfunctionsandpropertiesare

2、determinedbyitsstructureSoproteinstructurepredictionplaysanimportantruleunderstandingtherelationshipbetweenproteinstructureandfunctionandinthefieldsofpharmaceuticalsProteinsecondarystructurepredictionisthesignificantpart

3、ofproteinstructurepredictionanditplaysthebridgeroleandisacriticalstepinproteinstructurepredictionTraditionalmethodsofproteinstructureprediction，howeverhaveacoupleofproblems：lowpredictionaccuracydependentontheservertoolar

4、gebudgetspendingandusingonlyawaythatisnotcomplementarywithothermethods，whichseriouslyrestrictthedevelopmentofbioinformaticsMachinelearningisagoodmethodtopredictproteinsecondarystructureInthispaperthepredictionmodelusings

5、equencesalignmentsearchalgorithmcombinedartificialneuralnetworkisproposedtopredictproteinsecondarystructureFirst，Ukkonenalgorithmwithlowertimecomplexityisusedasthebasisofhomologoussequencesearchbecauseofthelargedataindat

6、abaseThen，inthemodelingofneuralnetworktopredictproteinstructure，inviewoftheslowconvergenceofneuralnetworksandeasilyfallingintolocalminimum，particleswarmoptimizationandsimulatedannealingmethodareusedtooptimizetheneuralnet

7、workAtlast，thehomologyinformationismadeuseofpredictingtheproteinstructureIntheexperiment，twocommonlyusedtestdatasetsareusedfortestingComparingwithcommonmethods，goodpredictionaccuracyisobtainedInaddition，theprogrambasedon

8、thepredictionmodelisindependentSOthatitdoesnotrelyonanyserverItisalsoeasytouse，offlineoperation，notrequiringalotofmoneyandhighperformancecomputersystemAlthoughthisapproachhasgoodpredictionresults，thepredictionaccuracyofs

9、omepartislowwhenfacingasmallamountoftestdatabecausethehomologyinformationcouldnotbeenfoundFinallythispaperputsforwardideastesolvethisproblembytheintegratedusingofavarietyofpredictingmethodsandtheadoptionofanintegratedapp