Enhancer-inclusive gene regulatory networks allow us to characterize and provide functional understanding of regulatory interactions underlying phenotypes in a cell-type specific manner.GRaNIE is a method that infers gene regulatory networks including enhancers from bulk transcriptomics (RNA-seq) and chromatin accessibility (ATAC-seq) data by reconstructing tripartite networks which describe transcription factor-enhancer-gene associations. Bulk data mask cell-type- and cell-state-specific gene expression and chromatin accessibility data as the counts represent an average of the activity of a population of cells. Hence, to overcome these limitations the approach requires its adjustment at the single-cell resolution to take into account inter-cell variation and consequently truly capture specific cell-type and cell-state regulatory differences. Therefore, a proper benchmark at the single-cell layer of the approach remains to be accomplished. We propose a way of benchmarking GRaNIE at the single-cell resolution through a validation-based approach that assesses the network's ability to predict real regulatory interactions from cell-type-specific promoter capture Hi-C, eQTL and ChIP-Seq data. Furthermore, we report further validation through a network biological comprehensive analysis alongside an evaluation of the eGRNs predictive power by means of GRaNPA, a machine learning framework that assesses the network's capability of predicting cell-type-specific differential gene expression data. The results show GRaNIE, applied to single-cell hiPSC-neuron timecourse datasets, is capable of inferring neuronal differentiation and development relevant regulatory associations as well as predicting cell-type specific iPSC against neuron DGE at the single-cell level.
- Preprocessing performed on the timecourse dataset.
- Preprocessing performed on the combined dataset.
- Extraction of raw data (RNA counts, ATAC counts, metadata) from the standardly preprocessed dataset-specific-seurat objects for further dataset-customized preprocessing.
- Conversion from seurat objects to AnnData objects to permit the usage of the preprocessed datasets for eGRN inference via SCENIC+ which is only available in Python/Scanpy.
- Get the ATAC fragments from the ArchR projects for each dataset.
- Run GRaNIE in batch mode (across a wide range of cluster resolutions on the integrated WNN space) generating an eGRN per resolution.
- Run GRaNIE in for a specific cluster resolution.
- Helper functions to execute GRaNIE for single-cell datasets. Functions to preprocess the data, run GRaNIE for a specific metadata column and in batch mode.
- Run GRaNIE pseudobulking the cells according a metadata column e.g. celltype.
- Run GRaNIE on the timecourse dataset for initial size tests, different cell type annotations and batch mode are utilized.
- Perform gene enrichment and TF enrichment analyses on a given eGRN.
- Customize GRaNIE original code that creates the peak-gene quality control plots in a qay that different eGRNs can be analyzed at the same time.
- Create a single pdf document with the final peak-gene quality control plots for different eGRNs (each network-specific QC plot in a different page)
- Run GRaNIE for separate scRNA and scATAC modalities for a NPC-Neuron datasets perturbed for a set of Schizophrenia markers.
- ChiP-seq data preprocessing.
- Validation of network TF-peak links with preprocessed ChiP-Seq data.
- Promoter capture Hi-C data preprocessing.
- Validation of network peak-gene links with preprocessed promoter capture Hi-C data.
- eQTL preprocessing and enrichment / validation analysis on network peak-gene links.
- Integration and visualization of networks validation.
- Networks evaluation for predicting cell-type-specific DGE using GRaNPA.
- Bulk RNA preprocessing and DGE analysis from an independent hiPSC to Neuron time course dataset.
- Visualization of a GRaNIE network and TF/Gene enichment analyses.
- Comparison of the effects of different cell-type annotations on GRaNIE networks.
- In depth study of pearson and spearman correlation algorithms for links inference in GRaNIE.
- GRaNIE networks analyses based on pseudobulking resolutions, overlaps on links across resolutions, networks general stats, pcHi-C validations, robust links and pcHI-C validated links relationhsip, and further pearson vs spearman correlation analysis.
- Overlap analysis of different type of links (TF-peak-gene, TF-peak and peak-gene) across cluster resolutions.
- Timecourse RNA preprocessing using Scanpy.
- Import the preprocessed timecourse Anndata objects and format it for further steps.
- Generate pseudobulk ATAC-seq profiles, call peaks and generate a consensus peak set for the timecourse dataset.
- Perform CisTopic modelling, candidate enhancer regions inference, motif enrichment analysis using pycisTarget for the timecourse dataset.
- eGRN inference using SCENIC+ on the timecourse dataset and further network analysis.
- Import the preprocessed combined Anndata objects and format it for further steps.
- Generate pseudobulk ATAC-seq profiles, call peaks and generate a consensus peak set for the combined dataset.
- Perform CisTopic modelling, candidate enhancer regions inference, motif enrichment analysis using pycisTarget for the combined dataset.
- eGRN inference using SCENIC+ on the combined dataset and further network analysis.
- eGRN inference using the Pando method.