|
|
大连理工大学学报 2014
ComputationalpredictionofMHCⅡ-peptideligandsbindingspecificitiesbyAUCOptimizedGibbsDOI: 10.7511/dllgxb201401005, PP. 28-36 Keywords: Gibbssamplingmethod,epitope,MHCⅡmolecules,reducedhomology Abstract: Inthedesignofpeptide-basedorotherdefinedantigen-basedvaccines,itisimportanttoknowwhichfragmentsofpathogen-derivedproteinswouldbindtotheMHCⅡmolecules.MoststudiesoftheMHCⅡepitopepredictionrarelygavethequantitativeanalysesofbindingspecificities.Sotheaccuracyofthesemodelsstillneedstobeimproved.AUCOptimizedGibbs(AOG)methodusesthehomologyreducedAUC,ratherthantherelativeentropytoguidethesampler.Itmakesboththepositiveandnegativeinformationofthesamplesbeincorporatedintothemodel.AOGachievesaverageAUCvaluesof0.771and0.713onthetenoriginalandhomologyreducedHLA-DR4(B1*0401)epitopebenchmarks,whicharebetterthan0.744and0.673bytheGibbssamplingmethod.InthequantitativeIEDBMHC-Ⅱbenchmarks,AOGachievesanaverageAUCvalueof0.766,comparedto0.718bytheTEPITOPE.AdetailedinspectionofinformationextractedfromHLA-DR4(B1*0401)dataallowstheidentificationofpositionswithobviousspecificities,i.e.,P1,P4,P6andP9positions,whichhavedistinctinfluenceontheMHC-peptidebinding.
|
