Predicting clinical outcome of neuroblastoma patients using an integrative network-based approach

Abstract

BackgroundOne of the main current challenges in computational biology is to make sense of the huge amounts of multidimensional experimental data that are being produced. For instance, large cohorts of patients are often screened using different high-throughput technologies, effectively producing multiple patient-specific molecular profiles for hundreds or thousands of patients.ResultsWe propose and implement a network-based method that integrates such patient omics data into Patient Similarity Networks. Topological features derived from these networks were then used to predict relevant clinical features. As part of the 2017 CAMDA challenge, we have successfully applied this strategy to a neuroblastoma dataset, consisting of genomic and transcriptomic data. In particular, we observe that models built on our network-based approach perform at least as well as state of the art models. We furthermore explore the effectiveness of various topological features and observe, for instance, that redundant centrality metrics can be combined to build more powerful models.ConclusionWe demonstrate that the networks inferred from omics data contain clinically relevant information and that patient clinical outcomes can be predicted using only network topological data.ReviewersThis article was reviewed by Yang-Yu Liu, Tomislav Smuc and Isabel Nepomuceno.Electronic supplementary materialThe online version of this article (10.1186/s13062-018-0214-9) contains supplementary material, which is available to authorized users.

A deep neural network approach to predicting clinical outcomes of neuroblastoma patients

“…We also note that the features extracted from the RNA-seq data are associated with lower performance than the equivalent features extracted from the microarray data. Both seems to contradict our previous study of the same classification problem, in which we reported no statistical difference between models built from both sets [25]. It is important to notice however that the learning algorithms and the data stratification are different between the two studies, which might explain this discrepancy.…”

“…We observe that the combined feature sets are not associated with any improvement upon the individual feature sets. This indicates that both sets might actually measure the same topological signal, which is in line with our previous observations [25]. Similarly, the integration of the data across the two expression datasets does not improve the average performance.…”

“…These reduced datasets are then used to infer Patient Similarity Networks (PSN), graphs in which a node represents a patient and an edge between two nodes represents the similarity between the two profiles of the corresponding patients. These graphs are built first, by computing the Pearson correlation coefficients between all profiles pairwise and second, by normalizing and rescaling these coefficients into positive edge weights through a WGCNA analysis [36], as described previously [25]. These graphs contain one node per patient, are fully connected and their weighted degree distributions follow a power law (i.e., scale-free graphs).…”

“…We then compare the DNN classifiers to other classifiers relying on different learning algorithms (SVM and RF). These algorithms have previously demonstrated their effectiveness to solve the same classification task on the 'Fischer ' dataset, albeit using a different patient stratification [37,25]. For a fair comparison, all classifiers are input the same features and are trained and tested using the same samples.…”

Cancer predictive studies

“…It is therefore essential, following a diagnosis of NB, to understand which molecular defect is present in each individual patient, in order characterize the patient’s risk group and select the more appropriate therapy, accordingly. In each of these groups, the identification of the underlying molecular events at the bases of tumour progression [ 59 – 61 ] might allow for specific combination therapies to optimise therapeutic efficacy and minimise toxic side effects [ 62 ].…”