DeepG4

DeepG4 predicts the probability that DNA sequences form active G-quadruplexes (G4s) in vitro and in vivo to analyze sequence determinants and variant effects on G4 activity.

Key Features:

• Prediction Accuracy: Demonstrates superior accuracy for predicting active G4s in cellular contexts compared with existing algorithms that focus on naked DNA.
• Convolutional Neural Network (Keras/TensorFlow): Employs a convolutional neural network architecture implemented with Keras and TensorFlow to recognize motifs associated with active G4 formation.
• Identification of Predictive Motifs: Identifies numerous specific sequence motifs predictive of G4 activity, including motifs corresponding to known transcription factors (TFs) that may influence G4 formation and stability.
• Cell Type-Specific Insights: Provides cell type–specific associations between TF motifs and G4 activity.
• Variant Impact Analysis: Predicts the effects of single nucleotide polymorphisms (SNPs) on G4 activity and downstream molecular features such as gene expression, H3K4me3 histone modification, and DNA methylation.

Scientific Applications:

• Genomic Mapping: Mapping active G-quadruplexes across genomes to study their roles in genomic stability and function.
• Transcriptional Regulation Studies: Linking TF-associated motifs and G4 activity to investigate regulatory mechanisms of gene expression.
• Disease Mechanism Exploration: Assessing how SNP-mediated changes in G4 activity could affect transcription, chromatin marks (e.g., H3K4me3), and DNA methylation for interpretation of disease-associated variants.

Methodology:

Convolutional neural network implemented in Keras and TensorFlow trained on DNA sequences to identify motifs predictive of active G-quadruplex formation and to output probabilities of G4 activity.