Fuzzy Logic Selection as a New Reliable Tool To Identify Gene Signatures in Breast Cancer – The INNODIAG Study
Magali Lacroix-Triki, Tatiana Kempowsky-Hamon, Carine Valle, Lyamine Hedjazi, Sophie Lamarre, Lidwine Trouilh, Laurence Puydenus, Loubna Mhamdi, Florence Dalenc, Thomas Filleron, Gilles Favre, Marie-Veronique Le Lann, Veronique Le Berre-Anton. Institut Claudius Regaud, Toulouse, France; CNRS, LAAS, Toulouse, France; INSA/CNRS/INRA - UMR5504/792, Toulouse, France
Background: For the last decade, breast cancer patient management has drastically evolved toward personalized medicine. In addition to the routinely used clinicopathological characteristics, clinicians will have to face an increasing amount of data derived from tumour molecular profiling. The aims of this study were to develop a new gene selection method based on fuzzy logic and classification algorithm, and to validate the gene signatures obtained with this original method on breast cancer patient cohorts.
Design: Based on fuzzy logic and classification algorithm, our selection method measures the contribution of each gene for each of two pre-defined classes in order to find the best discrimination. This algorithm extracts and ranks the most pertinent markers, since it is based on feature weighting according to optimal error rate, sensitivity and specificity. We applied the fuzzy logic selection on seven breast cancer microarray databases to obtain new gene signatures. To validate these gene signatures, we designed probes for the selected genes on Nimblegen custom microarrays and tested them on a series of 151 consecutive invasive breast carcinomas displaying clinicopathological features similar to those observed in routine practice.
Results: Using fuzzy logic selection, we identified new gene lists that were included in our Nimblegen in-house microarray (1100 genes,17000 probes). Analysis in the training public sets showed good performance of the 12 newly generated gene signatures for grade (sensitivity ranging from 80% to 100%, specificity 80% to 97%, error rate 3 to 12%). Similar results were obtained in our validation set of 151 breast cancer samples, with an error rate of less than 10% in the majority of the gene lists tested. Moreover, when applying in particular 4 distinct lists among the 68 histological grade III and 20 grade I tumours, 82 cases were correctly assigned in the two-class molecular grade. Interestingly, histological grade II tumours (n=63) were split in these two molecular grade categories with exception of 5 cases that showed an equivocal profile.
Conclusions: As a proof of concept, we confirmed the use of fuzzy logic selection as a new tool to identify gene signatures with good reliability and increased power of classification. This method based on artificial intelligence algorithms was successfully applied to molecular grade classification of breast cancers, but could be further developed for identification of any relevant prognostic or predictive multi-criteria signature in breast cancer.
Tuesday, March 5, 2013 8:30 AM
Proffered Papers: Section B, Tuesday Morning