Comparison of Different Sampling Algorithms for Phenotype Prediction

<p>On the Role of Artificial Intelligence in Genomics to Enhance Precision Medicine</p>

“…ð Þ<E tol f g , is composed by the sets of high predictive networks with similar predictive accuracy; that is, those sets of genes g that classify the samples with a prediction error O g ð Þ lower than E tol . [6][7][8] As in any other inverse problem, the cost function topography in phenotype prediction problems is composed of several flat curvilinear valleys 20,21 where the genetic signatures are located, all of them with a similar predictive accuracy of the training set. Nevertheless, in the phenotype prediction problem, the size of the high discriminatory genetic signatures varies, that is, high discriminatory genetic networks of different complexity exist, and the optimization of O g ð Þ is not always performed in the same space dimension.…”

“…[75][76][77] The important working hypothesis is that by sampling the uncertainty space of the phenotype prediction problems, we are able to understand the altered genetic pathways of the disease in order to use this knowledge in precision medicine for diagnosis, prognosis, and treatment optimization. Different interesting methods were proposed by Cernea et al 8 and successfully applied in the analysis of Triple Negative Breast Cancer metastasis, comparing the results obtained with Bayesian networks. 78 Bayesian networks are utilized to model the genetic signatures' distribution related to the phenotype prediction, P g=c obs À Á , according to Bayes' rule: [79][80][81] P g=c obs À Á , P g ð ÞP c obs =g À Á…”

“…We illustrated the importance of genomic robust sampling in precision medicine and uncertainty analysis with the analysis of the Triple Negative Breast Cancers (TNBC) phenotype. The algorithms utilized have been successfully utilized in the analysis of breast cancer and lymph node metastasis, 8,90 in Sarcopenia, 91 Multiple Sclerosis, 92 Multiple Myeloma 93 and Inclusion Body Myositis. 94 More specifically, we performed a robust sampling in order to find out the altered genetic pathways by the metastasis events.…”

“…Nowadays, there are available tools that are capable of collecting a large amount of genomic data, alongside with cutting-edge data analytics for interpretation, which aid in our understanding of genomics, disease mechanisms, and treatments ( Figure 1). [8][9][10]…”

Robust Sampling of Defective Pathways in Multiple Myeloma

“…The defective pathways are sampled via two different algorithms: Fisher’s ratio [4] and holdout [5] samplers. These algorithms have been used to unravel the altered pathways involved in the metastasis in triple negative breast cancer outperforming Bayesian networks [6]—and recently in Parkinson disease [7] —to provide new insights about the defective pathways which are involved. The problem addressed in this paper does not consist in just solving the classification problem involved in phenotype prediction, but in finding the genetic pathways that are involved in the genesis and development of this disease, which is hampered by the high degree of under-determinacy of these kind of problems.…”