Identification of three-genomic regions prognostic signature in small node-negative breast carcinomas.

Abstract

#1081 Background: The purpose of this study was to identify a genomic signature of early metastatic recurrence, in order to predict accurately breast carcinomas clinical outcome and to select patients with node negative and small tumor size (<3cm) who would benefit from adjuvant chemotherapy.
 Patients and methods: Using genome-wide BAC-PAC Genomic Comparative Hybridization (CGH) array (1kb), we analyzed a training set of 78 patients. All patients had invasive ductal carcinomas and were initially treated by surgery and radiotherapy, without chemotherapy. The validation was performed on an independent test set of 90 patients. The training and tests sets were composed of respectively 53 and 58 patients disease-free survivors at 60 months (good prognosis group), and by 25 and 32 patients with distant metastatic recurrence before 48 months (poor prognosis group). In the training set, a signature was established as a logistic multivariate model of regions containing contiguous BAC clones with statistically different ratios and median frequencies of gains and losses between the poor and the good prognosis groups. This signature was then validated using the independent test set to evaluate its accuracy to classify T0T1T2N0 patients according to their outcome.
 Results: The training test identified a prognostic signature defined by 3 genomic regions, located on the 2p (38.3 to 40.9Mb), 3p (32 to 80.3Mb), and 8q (78.8 to 128.9Mb) chromosomes. In the test set, 90% of patients of favourable outcome were ER +ve and 88% were PR +ve, compared to 62% and 55% in the poor outcome group, respectively. In the test set, our signature was highly informative to identify patients that developed distant metastases before 48 months: the rate of patients well classified was 0.74, CI (95%): [0.64; 0.83], with a specificity of 95%, CI (95%): [86%; 99%]. On Kaplan-Meier analysis, the poor-prognosis genomic signature group of patients had a RR of 3.5 of metastatic relapse (log rank test p<0.001).
 Conclusions: Our signature, validated on an independent series of small T0T1T2N0 and on a majority of ER/PR positive tumors, may provide a robust and accurate tool to identify, in addition to classical parameters, patients who would benefit from adjuvant medical treatments. The comparison of this genomic signature with RNA based signatures and clinico-pathological parameters, is currently being investigated. Citation Information: Cancer Res 2009;69(2 Suppl):Abstract nr 1081.

Classifying Microarray Gene Data of Various Cancerous Tissues Using Metric Learning

“…Gen Say. Alt Tür Sayısı Gravier [19] Meme Kanseri 168 2905 2 Sorlie [20] Meme Kanseri 85 456 5 Pomeroy [21] ÇÇBT 60 7128 2 Golub [22] Lösemi 72 7129 2 Shipp [23] Lenfoma 77 6817 2 Khan [24] KYMHT 63 2308 4…”

“…Gerçekten, incelenen 6 veri seti için LMNN sınıfları ayırmada başarılı olmuştur. Bu anlamda LMNN'yi ITML metodu takip etmiş, Gravier [19] veri seti hariç diğer veri setlerinde sınıfları ayrıştırabilmiştir. En son olarak NCA'nın da Golub [22] ve Shipp [23] veri setlerinde ayrı sınıflara ait örnekleri birbirinden uzaklaştırmada başarılı olduğu söylenebilir.…”

“…EYMS'nin kanser alt türünü tahmin etme başarısı ise k-NN'ye göre daha dar bir aralıkta değişkenlik gösterir; yaklaşık %74 -%89 bandında değişir. %74 sınıflandırma başarısının elde edildiği Gravier[19] veri setinde Khan[23] veri setinde olduğu gibi farklı sınıflara ait örnekler birbirleriyle iç içe geçmiştir (bakınız Şekil 1a). 𝑘-NN gibi bir lineer sınıflandırıcı olan EYMS'nin sınıfların bu şekilde iç içe olduğu veri setlerinde görece başarız sonuçlar alması bu yüzden şaşırtıcı değildir.…”

“…Farklı sınıflara ait örneklerin lineer bir doğru ile birbirinden ayrılamadığı uzaylarda sınıflandırma başarısı düşük kalır. Gravier[19] veri setindeki sınıf tahminin yeterince başarılı olmaması bu şekilde açıklanabilir. Özel olarak LVQ sınıflayıcısı Pomeroy[21] veri setinde en düşük sınıflandırma başarısına ulaşmıştır (%58).…”