NR ATLS
AU Howells,L.; Horton,R.A.; Jackman,R.; Everest,S.J.; Allison,G.G.; Moorby,J.M.
TI The use of pattern recognition approaches for the identification of BSE in cattle from patterns of redox metabolites in blood plasma.
QU International Conference - Prion 2005: Between fundamentals and society's needs - 19.10.-21.10.2005, Congress Center Düsseldorf - Poster Session: Diagnosis DIA-56
PT Konferenz-Poster
AB
To study changes in blood composition during the incubation of BSE, plasma samples from 10 experimentally challenged cows were collected from 763 to 1743 days post inoculation. These, together with samples collected from 10 healthy control cows over the same period, were analysed by HPLC with electrochemical detection to produce a pattern or "fingerprint" of electrochemically active compounds. The data from challenged and control cows were compared to determine differences in the metabolite fingerprints.
In this work we use the pattern recognition methods principle components analysis (PCA), SIMCA (soft independent modeling of class analogy) and Gmax to analyse the complex three-dimensional metabolomic profile obtained from our determination of redox analytes in bovine plasma following separation by liquid chromatography.
PCA of the data showed most of the variance was explained in the first principle component (94.4%) with 2.7, 1.5, 0.66, 0.27 and 0.12 in PCs 2, 3, 4, 5 and 6 respectively. "Scores" and "loadings" were plotted but these gave no clues of correlations with the BSE state of the animals.
SIMCA cross-validates the PCA model of each class using "leave-one-out" cross-validation. The application of this algorithm to our data using autoscaled variables had some predictive power and the algorithm was able to correctly categorise all 72 plasma samples.
The models created by TheGmax program were unable to assign class with 100% accuracy. The best ones produced a single misclassification in each group. ie: gave 2.63% false negatives and 2.94% false positives.
Using both the GMax and SIMCA we have identified a range of variables with the theoretical ability to discriminate the two sample groups. We have also demonstrated that SIMCA can be used to determine the disease state of individual animals from the differences in the patterns of the variables in the totality of each blood plasma sample.
AD L.Howells, R.A.Horton, R.Jackman, S.J.Everest, VLA-Weybridge, Addlestone, Surrey, UK; G.G.Allison, J.M.Moorby, Institute of Grassland and Environmental Research, Aberystwyth, UK
SP englisch
PO Deutschland