A multiobjective deep learning approach for predictive classification in Neuroblastoma

  • 2018-02-22 18:43:29
  • Valerio Maggio, Marco Chierici, Giuseppe Jurman, Cesare Furlanello
  • 0

Abstract

Neuroblastoma is a strongly heterogeneous cancer with very diverse clinicalcourses that may vary from spontaneous regression to fatal progression; anaccurate patient's risk estimation at diagnosis is essential to designappropriate tumor treatment strategies. Neuroblastoma is a paradigm diseasewhere different diagnostic and prognostic endpoints should be predicted fromcommon molecular and clinical information, with increasing complexity, as shownin the FDA MAQC-II study. Here we introduce the novel multiobjective deeplearning architecture CDRP (Concatenated Diagnostic Relapse Prognostic)composed by 8 layers to obtain a combined diagnostic and prognostic predictionfrom high-throughput transcriptomics data. Two distinct loss functions areoptimized for the Event Free Survival (EFS) and Overall Survival (OS)prognosis, respectively. We use the High-Risk (HR) diagnostic information as anadditional input generated by an autoencoder embedding. The latter is used asnetwork regulariser, based on a clinical algorithm commonly adopted forstratifying patients from cancer stage, age at insurgence of disease, and MYCN,the specific molecular marker. The architecture was applied to IlluminaHiSeq2000 RNA-Seq for 498 neuroblastoma patients (176 at high risk) from theSequencing Quality Control (SEQC) study, obtaining state-of-art on thediagnostic endpoint and improving prediction of prognosis over the HR cohort.

 

Quick Read (beta)

loading the full paper ...