Deep Learning/Artificial Intelligence and Blood-Based DNA Epigenomic Prediction of Cerebral Palsy

Int J Mol Sci. 2019 Apr 27;20(9):2075. doi: 10.3390/ijms20092075.

Abstract

The etiology of cerebral palsy (CP) is complex and remains inadequately understood. Early detection of CP is an important clinical objective as this improves long term outcomes. We performed genome-wide DNA methylation analysis to identify epigenomic predictors of CP in newborns and to investigate disease pathogenesis. Methylation analysis of newborn blood DNA using an Illumina HumanMethylation450K array was performed in 23 CP cases and 21 unaffected controls. There were 230 significantly differentially-methylated CpG loci in 258 genes. Each locus had at least 2.0-fold change in methylation in CP versus controls with a FDR p-value ≤ 0.05. Methylation level for each CpG locus had an area under the receiver operating curve (AUC) ≥ 0.75 for CP detection. Using Artificial Intelligence (AI) platforms/Machine Learning (ML) analysis, CpG methylation levels in a combination of 230 significantly differentially-methylated CpG loci in 258 genes had a 95% sensitivity and 94.4% specificity for newborn prediction of CP. Using pathway analysis, multiple canonical pathways plausibly linked to neuronal function were over-represented. Altered biological processes and functions included: neuromotor damage, malformation of major brain structures, brain growth, neuroprotection, neuronal development and de-differentiation, and cranial sensory neuron development. In conclusion, blood leucocyte epigenetic changes analyzed using AI/ML techniques appeared to accurately predict CP and provided plausible mechanistic information on CP pathogenesis.

Keywords: DNA methylation; cerebral palsy; epigenetics; neurodegenerative disorders; newborns.

MeSH terms

  • Artificial Intelligence*
  • Case-Control Studies
  • Cell-Free Nucleic Acids*
  • Cerebral Palsy / blood
  • Cerebral Palsy / genetics*
  • Cerebral Palsy / metabolism
  • CpG Islands
  • DNA Methylation
  • Deep Learning*
  • Epigenesis, Genetic*
  • Epigenomics / methods
  • Gene Expression Profiling
  • Gene Regulatory Networks
  • Humans
  • Infant, Newborn
  • ROC Curve

Substances

  • Cell-Free Nucleic Acids