Article Figures & Data
Figures
- Fig. 1.
Chromosomal aberrations. The percentage of tumors showing gain or loss at a specific chromosomal location (evaluated at the channel level) for the BRCA1 group (N = 28, black line) and the control group (N = 42, gray line). Gray shaded, difference between the groups. Significant arms (χ2; P < 0.05) are listed in Table 2 <$REFLINK> .
- Fig. 2.
Cytogenetic banding and 0.85VAF banding of chromosome 3 and 5. Top, statistically significant (adjusted P < 0.1) chromosomal regions for the mutual information criterion indicated as gray shaded areas for the arm (only 3q is shown and not 10p), 0.85VAF band, and channel resolution. Bottom, CGH ratio profiles after classification of the 28 BRCA1 and 42 control tumors using the BRCA1 classifier based on the 0.85VAF banding and discrete representation. Ten controls were misclassified as BRCA1 (bottom). Two horizontal lines at ratios of 0.85 and 1.15 delineate the range associated with no genetic change. Vertical line, positions of centromeres.
- Fig. 3.
A posteriori densities for the different regions for the best classifier. The distributions for the selected regions, bands 3.1, 3.5 and 5.2 (0.85VAF) are depicted in A, B, and C, respectively. The a posteriori density histograms represent the probability that a tumor is BRCA1 or control given the particular aberration observed in the tumor. For example, for a tumor with a gain of band 3.1, the a posteriori distribution (A) indicates that the probability that the tumor belongs to the BRCA1 class is high (0.75) and that it is low for the control class (0.25).
- Fig. 4.
ROC. The ROC for the best classifier (discrete SBC on selected 0.85VAF regions) is given. The curve depicts, for every combination of prior probabilities of the BRCA1 and control groups, the positive rate and the negative rate. The positive rate is defined as the number of correctly classified BRCA1 tumors; the negative rate represents the number of control tumors classified as BRCA1 tumors. ○, position of the optimal classifier on the curve.
- Fig. 5.
BRCA1 classification score. For each individual tumor of the training set and the validation set (BRCA1 and control), the BRCA1 classification score as calculated using the simple Bayes rule, is depicted. A high score indicates a high probability to belong to the BRCA1 class. The percentage of tumors from each group that are classified to the BRCA1 class is indicated.
Tables
- Table 1
Clinicopathological characteristics of breast carcinomas from BRCA1 mutation carriers in comparison to control cancers
BRCA1 Control Clinical characteristics No. of tumors analyzed 28 42 Patients with breast and ovarian carcinoma 3 9 Patients with bilateral breast carcinomas 12 22 Pairs of bilateral breast carcinomas analyzed 5 13 Mean age at diagnosis (range) 40 (28–62) 51 (31–73) Histology Invasive ductal carcinoma 28 35 Invasive lobular carcinoma 0 2 Other 0 5 Grade I 0 3 Grade II 6 20 Grade III 22 19 Immunohistochemistry ER +a 12% (3/24) 57% (21/37) PR+ 17% (4/23) 44% (16/36) Neu + 0% (0/22) 11% (4/35) Cyclin D1 + 0% (0/13) 15% (5/34) P53 + 41% (9/22) 26% (9/34) -
a ER, estrogen receptor; PR, progesterone receptor; Neu, HER2-neu; +, positive.
-
- Table 2
Significant chromosomal regions between BRCA1 and control tumors
Percentage of tumors from BRCA1 and control tumors with loss or gain of chromosome arms. Only chromosome arms with a significant difference (χ2, not corrected for multiple sampling, P < 0.05) between the groups are listed.
Loss BRCA1 (%) Control (%) P Gain BRCA1 (%) Control (%) P 3p 61 31 0.014 3q 79 26 0.000 4p 64 38 0.032 7p 39 12 0.008 5q 82 40 0.001 8q 96 67 0.003 12q 54 26 0.020 10p 46 14 0.003 16p 32 12 0.038 12p 46 17 0.007 18q 54 26 0.020 16p 11 31 0.048 17q 29 52 0.049 - Table 3
Significant regions at different levels of resolution
Summary of the chromosomal regions where the filtering criteria (mutual information and U test) detected significant (adjusted for multiple sampling, P < 0.1) correlations with the class labels (BRCA1 and control).
Total no. of regions Discrete (Mutual information) Continuous (U test) Region Adjusted P Region Adjusted P Arm 41 3q 0.013 Not applicable 10p 0.080 Channel 1716 5.26–5.30 <0.10 3.83–3.98 <0.10 5.27–5.50 <0.10 Band 0.85VAF 81 3.1 0.015 3.2 0.053 3.5 0.002 3.5 0.005 5.2 0.009 5.2 0.003 5.3 0.078 - Table 4
Performance of BRCA1 classifiers
Summary of the best LOOCV performance and associated region sets for classification of BRCA1 tumors. Results are shown for the wrapped discrete SBC, as well as for the mutual information filtered SBC for both discrete and continuous representations. Only region sets where limited variation occurred in the selected sets (across the LOOCV loops) are listed.
Wrapped discrete SBC Mutation information filtered SBC Regions Performance (%) Regions Performance (%) Discrete representation Arms 3q and 10p 79 3q and 10p 73 Bandsa 3.5 and 5.2 and (2.1 or 10.2 or 3.1) 77 3.1 and 3.5 and 5.2 84 Channels 11 regions 64 16 regions 67 Continuous representation Arms Not applicable Not applicable Bandsa 9 regions 67 3.5 and 5.2 76 Channels Too computationally expensive 5.31 and 5.32 77 -
a 0.85VAF bands.
-
- Table 5
False positive analysis
Profile, classifier scores, and clinical parameters for the false positives and false negatives identified by the best classifier.
Tumor Features 3.1, 3.5, 5.2 Classifier score Age (1st, 2nd) Ovarian tumor Histological grade PR status ER status Tumor type False positives 1 [G,G,N]a 0.95 58 No 3 − − IDC 2 [G,G,N] 0.95 45,47 Yes 3 − − IDC 3 [G,G,N] 0.95 45,47 Yes 3 − − MP 4 [L,G,N] 0.93 33,39 No 3 ND ND IDC 5 [L,G,N] 0.93 52,52 No 2 + + ILC 6 [N,G,N] 0.65 49,61 Yes 3 − + IDC 7 [N,G,N] 0.65 48,53 No 3 − − MP 8 [L,N,N] 0.61 39,42 No 3 − + IDC 9 [L,N,N] 0.61 44,57 No 3 − − IDC 10 [L,N,N] 0.61 45,54 No 2 + − IDC False negatives 11 [N,N,N] 0.17 35,46 No 2 + + IDC -
a [G,N,L], gain, normal, loss; ND, not done; ER, immunohistochemistry for estrogen receptor; PR, immunohistochemistry for progesterone receptor; IDC, invasive ductal carcinoma; MP, metaplastic; ILC, invasive lobular carcinoma; Grade, histological grade.
-
Volume 62, Issue 23
