A methodology for detecting relevant single nucleotide polymorphism in prostate cancer with multivariate adaptive regression splines and backpropagation artificial neural networks

• Published in: Neural computing & applications Vol. 32; no. 5; pp. 1231 - 1238
• Main Authors: Sánchez Lasheras, Juan Enrique, González Donquiles, Carmen, García Nieto, Paulino José, Jiménez Moleon, José Juan, Salas, Dolores, Suárez Gómez, Sergio Luis, de la Torre, Antonio J. Molina, González-Nuevo, Joaquín, Bonavera, Laura, Carballido Landeira, Jorge, de Cos Juez, Francisco Javier
• Format: Journal Article
• Language: English
• Published: London Springer London 01.03.2020; Springer Nature B.V
• Subjects: Artificial Intelligence; Artificial neural networks; Back propagation; Computational Biology/Bioinformatics; Computational Science and Engineering; Computer Science; Data Mining and Knowledge Discovery; Image Processing and Computer Vision; Multivariate analysis; Neural networks; Nucleotides; Polymorphism; Probability and Statistics in Computer Science; Prostate cancer; Proteins; Regression analysis; S.I. : AI and ML applied to Health Sciences (MLHS); Spline functions; Support vector machines; Single nucleotide polymorphism; Multivariate adaptive regression splines; Prostate cancer; Artificial neural networks
• ISSN: 0941-0643; 1433-3058; 1433-3058
• DOI: 10.1007/s00521-018-3503-4

The objective of the present paper is to model the genetic influence on prostate cancer with multivariate adaptive regression splines (MARS) and artificial neural networks (ANNs) techniques for classification. These models will be able to classify subjects that have cancer according to the values of the selected proteins from the genes selected with the models as most relevant. Subjects are selected as cases and controls from the MCC-Spain database and represent a heterogeneous group. Multivariate adaptive regression splines models allow to select a set of the most relevant proteins from the database. These models were trained in nine different degrees and chosen regarding its performance and complexity. Artificial neural networks models were trained on with data restricted to the most significant variables. The performance of both types of models was analyzed in terms of the area under the curve of the receiver operating characteristics curve. The ANN technique resulted in a model with AUC of 0.62006, while for MARS technique, the value was 0.569312 in the best situation. Then, the artificial neural network model obtained can determine whether a patient suffers prostate cancer significantly better than MARS models and with high rate of success. The best model presented was based on support vector machines, reaching values of AUC of 0.65212.