摘要
Anewmethodwasdescribedforusingarecurrentneuralnetworkwithbiasunitstopredictcontactmapsinproteins.Themaininputstotheneuralnetworkincluderesiduespairwise,residueclassificationaccordingtohydrophobicity,polar,acidic,basicandsecondarystructureinformationandresidueseparationbetweentworesidues.Inourwork,adatasetwasusedwhichwascomposedof53globulinproteinsofknown3Dstructure.Anaveragepredictiveaccuracyof0.29wasobtained.Ourresultsdemonstratetheviabilityoftheapproachforpredictingcontactmaps.
出版日期
2005年03月13日(中国期刊网平台首次上网日期,不代表论文的发表时间)