A new method of DNA probes selection and its use with multi-objective neural network for predicting the outcome of breast cancer preoperative chemotherapy.

DNA microarrays technology has emerged as a major tool to explore cancer biology and solve clinical issues. The response to chemother- apy represents such an issue because its prediction would make it possible to give the patients the most appropriate chemotherapy regimen. We pro- pose a new method of probes selection, and we study the performances of predictors designed with multi-objective neural network (MOBJ-NN) taking as input the expression levels of the selected probes. The novelty of this paper is to link the method of probes selection and the MOBJ-NN model for designing multi-gene predictors. The development of post-genomic high-throughput measurement technologies and the associated computational analysis tools give the opportunity to iden- tify for each tumor, a profile based on level of mRNA expression. In breast cancer, neoadjuvant chemotherapy (treatment given prior to surgery) makes it possible to check, in vivo, breast tumor chemosensitivity. A pathologic com- plete response (PCR) at surgery is correlated with an excellent outcome while residual disease (NoPCR) is associated with a poor outcome. An accurate pre- diction of tumor sensitivity to preoperative chemotherapy is an important issue because patients with predicted residual disease may avoid the prescription of an inefficient treatment and may be allocated to other treatments. The design of multigene predictors of the patients' class, PCR or NoPCR, is a supervised learning problem. The methods that are the most commonly used for selecting a subset of DNA probes are based on the identification of probes that depart the most from a random distribution of expression levels. The DNA probes are ranked and the genes are selected according their p-values of a t-test. In these methods, the forthcoming classifier models are not involved in the selection pro- cess (such mathod are often said to be part of a 'filtering approach'). In some other studies the classifier model is involved in the process of DNA probes se-