Script for scTenifold

[marinaEM]

Contribution Guidelines

1. Initial Setup

   • Clone the repository: git clone <repository_url>

2. Update Local Content

   • Ensure you’re up to date with the development branch:

     git checkout devel
     git pull origin devel

3. Create a New Branch

   • Create a descriptive branch for your work:

     git checkout -b <branch_name>

4. Commit & Push Changes

   • Stage and commit your changes with clear messages:

     git add <file_name>
     git commit -m "Brief description of changes"

   • Push to GitHub:

     git push origin <branch_name>

5. Create a Pull Request

   • Target the devel branch for merging.
   • Follow the PR template, review, and submit.

6. Final Merge & Update

   • Once approved, merge your PR and delete your branch if desired.
   • Update your local repository:

     git checkout devel
     git pull origin devel

By following these steps, you contribute effectively and collaboratively.

Task: Implement script for scTenifold

• Description: Create a script or set of scripts to run scTenifold, defining input data and output format. This should prepare the tool's output for integration into Nextflow.
• Input: Dataset data, metadata, and perturbation task
• Output: Prediction results from scTenifold

Tags: #python, #GRNs

[guzikine]

Short tool description: scTenifoldKnk utilizes expression data from scRNA-seq of the WT samples as input and constructs a denoised single-cell GRN (scGRN) which can be then used for KO analysis. KO gene(s) is then analysed using enrichment analysis (GSEA).

Advantages of the tool:

• Can perform simultaneous knockouts of multiple genes, allowing for the exploration of gene interactions and epistatic effects in a single analysis
• The tool is species-agnostic, meaning it can be applied to scRNA-seq data from various organisms
• Allows for the definition (as a parameter) of maximum ratio of mitochondrial reads (mitochondrial reads / library size) present in a cell to be included in the analysis
• Incorporates tensor decomposition techniques to denoise the adjacency matrix, which helps to filter out noise from the scRNA-seq data
• Uses GSEA as the enrichment method which is more advantageous than for example ORA, because it shows which genes are down- or up-regulated
• The R package provides a single function which is used to perform the whole KO analysis, which is very convenient, especially for researchers with limited computational background

Disadvantages of the tool:

• The tool uses principal component regression which assumes linear relationships between genes, which may not capture the full complexity of gene interactions, especially in nonlinear biological systems (contrast for tools like DeepReg or GRNBoost)
• The tool only uses it's own generated gene regulatory networks, which, if poorly defined or incomplete, may lead to less reliable predictions (incorporation of known biological networks as a validation for the newly constructed GRN might be a good idea)
• The accuracy of predictions is highly dependent on the quality and completeness of the input scRNA-seq data (as a contrast, tools like CellOracle incorporates additional data integration techniques to enhance robustness against data quality issues)
• Complex output data structure poses challenges with interpretation and further analyses utilising this data object
• The parameters for the main function are not intuitive and requires advanced knowledge in linear algebra/graph theory to be able to modify them as well as understand the parameters
• No clear tutorial or vignette which makes the tool unattractive

[guzikine]

Input. A single scRNA-seq measurement matrix with barcoded cells in the columns and genes in rows (mitochondrial genes should start with MT- prefix). Parameter gKO defines which gene to knockout, it can also be a vector of genes (more than one), i.e. scTenifoldKnk(..., gKO = c("Gene1", "Gene2"))

Output. The function scTenifoldKnk() outputs a data object with three lists:

• tensorNetworks - this list contrains weight matrices for the WT and KO networks
• manifoldAlignment - this list contains the manifold alignments for the WT and KO networks, which contains values that represent the coordinates or positions of genes in a reduced-dimensional space after aligning the WT (NLMA 1 column) and KO (NLMA 2 column) GRNs.
• diffRegulation - this list contains information about each gene after differential gene expression analysis comparing both GRNs. It contains Euclidian distance, fold-change, p-value, Z-score values.