60 likes | 218 Views
Figure 1. Cluster analysis. DIGE analysis. 2. Prediction analysis of microarray (PAM). 1. 2d hierarchical clustering heatmap plotting. DIGE raw gel images SJIA (13 F, 13 Q) POLY (5 F, 5 Q). Predictor discovery in training set. Predictor test in testing set. 6. 7. 3.
E N D
Figure 1 Cluster analysis DIGE analysis 2 Prediction analysis of microarray (PAM) 1 2d hierarchical clustering heatmap plotting DIGE raw gel images SJIA (13 F, 13 Q) POLY (5 F, 5 Q) Predictor discovery in training set Predictor test in testing set 6 7 3 Spot finding spot alignment feature extraction Training set SJIA (24 F, 14 Q) POLY (15 F, 10 Q) Testing set SJIA (24 F, 14 Q) POLY (15 F, 11 Q) Mann Whitney P < 10-5 10 protein candidates Normalization manual review 8 Classifier training PAM class prediction algorithm 7 feature biomarker panel 4 Predictors of 4 ~ 12 features 889 discrete spot features Literature review + literature candidates - check antibody availability Ten-fold cross-validation LDA Manual review Prospective Study N.D. Discriminate SJIA F KD FI Classify SJIA F vs Q POLY F vs Q MSMS ID 97 spots Analysis of goodness of class separation 5 12 ELISA assays
SJIA SJIA F F Q Q SJIA SJIA POLY POLY F F Q Q F F Q Q Figure 2 A B C D SAP A2M APOAIV CFHR1 HP ATIII CRP ATIII HP GSN A2M A2M CFHR1 GSN GSN TTR GSN APOA1 A2M APOA1 ATIII APOA1 ATIII APOA1 TTR APOA1 APOA1 APOA1 APOA1 SAP APOIV APOIV APOA1 HP APOA1 CRP APOA1 HP MRP14 HP HP HP HP HP HP HP HP HP HP MRP8 MRP8 MRP8 MRP8 MRP14 SAA SAA SAA SAA SAA SAA
Figure 3 A B C D Goodness of class separation – D probability POLY Training POLY Testing SJIA Training SJIA Testing Feature# 4 5 7 8 12 4 5 7 8 12 4 5 6 7 8 9 12 4 5 6 7 8 9 12
Figure 4 TRAINING TESTING TRAINING TESTING A B C D POLY Q SJIA Q POLY F SJIA F POLY Q SJIA Q POLY F SJIA F Estimated probability Samples
Figure 5 B A C D Training set n = 38 Testing set n = 38 All data n = 76 SJIA • Biomarker panel • of 7 members • HP • APO AI • A2M • SAP • CRP • MRP8/MRP14 • SAA SJIA SJIA Clinical diagnosis Clinical diagnosis Clinical diagnosis F Q F Q F Q 24 14 24 14 48 28 n = n = n = LDA LDA LDA Classified as F Classified as F Classified as F 18 5 21 4 43 9 Classified as Q Classified as Q Classified as Q 6 9 3 10 5 19 75% 64.3% 87.5% 71.4% 89.6% 67.9% Percent Agreement with clinical diagnosis + - Percent Agreement with clinical diagnosis + - Percent Agreement with clinical diagnosis + - 71% 81.5% 81.6% Overall P = 3.6 X 10-2 Overall P = 3.9 X 10-4 Overall P= 3.3 X 10-7
SAP Figure 6 A B C D SJIA F SJIA F KD KD FI FI FI KD SJIA F Data set n = 37 Data set n = 88 48 13 10 12 12 30 SJIA F SJIA F NOT-SJIA F NOT-SJIA F Clinical diagnosis Clinical diagnosis 48 13 40 24 n = n = SAA MRP8 PAM Unsupervised clustering MRP14 HP Classify as SJIA F Clustered as SJIA F 10 48 3 1 Clustered as NOT-SJIA F Classify as NOT-SJIA F 3 0 37 23 CRP 100% 77% 92.5% 96% - - Percent Agreement with clinical diagnosis Percent Agreement with clinical diagnosis + + APOA1 96.5% 85% Overall P= 8.12 X 10-6 Overall P= 2.2 X 10-16 A2M