Figure 1
From: A neural network based model effectively predicts enhancers from clinical ATAC-seq samples
Summary of the PEAS framework. Features were extracted from ATAC-seq bam files and genomic sequences for each OCR. OCRs were described based on 24 data features and labeled using ChromHMM states (step 1). Classification models were built in 5 cell types (GM12878, PBMC, CD4+ T, CD14+ monocytes and islets) using MLP neural networks (step 2). We built a combined model for predictions in cell types without annotations by pooling data across five different cell types (step 3), which we applied on EndoC-βH1, naïve CD8+ T, K562, and MCF7 ATAC-seq data. Individuals’ enhancers were predicted from islet ATAC-seq samples (n = 19) using a model trained in islet cells (step 4).
