Modeling cancer outcome prediction by aiNet: Discrete artificial immune network

The paper proposes a methodology of using artificial immune networks (aiNets) for a classification problem oriented to cancer outcome prediction on the basis of gene expression profiling. The aiNet is a discrete (iterative) computational model that adopts clonal selection and affinity maturation principles from biological immune systems. The treatment (chemotherapy or stem cell support) of diffuse large B-cell lymphoma (DLBCL) depends on the distinction of the significant subtypes of this kind of cancer and influences on the cured/fatal outcome of the disease. Two aiNets discover clusters in the two classes - cured and fatal patients. After learning the memory cells of the aiNets represent the internal images (in a compressed form) of the DLBCLs presented as training samples. The cancer outcome predictor, based on the two aiNets, is tested in MATLAB environment on 58 data samples (32 cured and 26 fatal) available in the literature. A high prognostic accuracy is achieved by a 13-gene outcome prediction model.