Regensburg 2000 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Downloads | Hilfe
DY: Dynamik und Statistische Physik
DY 46: POSTER II
DY 46.63: Poster
Donnerstag, 30. März 2000, 15:00–18:00, D
Analyzing peptide maps with neural nets — •H. Broll1, Ch. Ziegaus1, Ch. Bauer1, E.W. Lang1, and M. Woznyi2 — 1Institut für Biophysik und physikalische Biochemie, Universität Regensburg, 93040 Regensburg — 2Roche GmbH, Therapeutics, Nonnenwald 2, D-82372 Penzberg
The analysis of processed proteins is routinely done with HPLC methods. The resulting peptide maps provide a finger print characteristic of any processed protein and allow abnormal protein fragments to be detected. Currently the analysis of peptide maps has still to be done by an expert via visual inspection. A largely automatized analysis seems highly desirable. In this investigation we study the use of artifical neural nets to analyse and classify such peptide maps. First, highly non-linear artifactual run-time shifts between a standard and any observed spectrum due to changing conditions on the HPLC columns are learned by a radial basis function (RBF) network. The latter then provides the means to correct any spectrum to be analyzed. PCA algorithms are then used to reduce the dimensionality of the input considerably and to extract significant features from the spectra. These feature vectors are then further analyzed by self-organizing maps to cluster the input and classify the spectra into two classes: well and badly processed proteins.