Integration of GRN inference algorithms using single cell multi-omics data to BEELINE

[yiqisu]

Background

Gene regulatory networks (GRNs), which delineate the interactions between transcription factors (TFs) and their target genes, represent the mechanisms governing cell differentiation and disease development. Our benchmarking framework, BEELINE [1], rigorously assessed a dozen unsupervised algorithms for GRN inference in early 2020. Since then, with the rapid advances in single-cell sequencing technologies, a plethora of algorithms for GRN inference have emerged over the past four years. To accommodate newly released algorithms and datasets, we plan to introduce an enhanced framework, BEELINE 2.0, aimed at offering users valuable recommendations for the selection or development of GRN inference algorithms.

Reference: [1] Pratapa, A., Jalihal, A.P., Law, J.N. et al. Benchmarking algorithms for gene regulatory network inference from single-cell transcriptomic data. Nat Methods 17, 147–154 (2020). https://doi.org/10.1038/s41592-019-0690-6

Goal

The objective of this project is to extend the BEELINE pipeline by incorporating novel GRN inference algorithms, particularly those developed for using single-cell multi-omics data (scRNA-seq and scATAC-seq data). We are interested in evaluating both the performance of new GRN inference algorithms as well as how the integration of two types of information contributes to GRN inference. The key steps in the project will include:

1. Read the BEELINE paper to understand the benchmark.

2. Install BEELINE or clone the repository and run it to replicate results in the paper.

3. Integrate two or all three GRN inference algorithms from the following list to BEELINE.

   • DeepMAPS (based on a multi-head graph transformer): [Paper link] [GitHub repo]
   • scTIE (using an autoencoder): [Paper link] [GitHub repo]
   • MICA (a non-linear method based on mutual information): [Paper link] [GitHub repo]

4. Integrate these three selected algorithms to BEELINE extension pipeline with the following steps:

   • Refer to the instructions on adding an algorithm to BEELINE. In addition, the README.md in each algorithm’s folder contains a specialized version of how we integrated that particular method into BEELINE (see https://github.com/Murali-group/Beeline/tree/master/Algorithms).
   • The student will have to expand the general instructions as needed based on additional steps required by the newly-added algorithms.

5. The mentor will provide appropriate scRNA-seq and scATAC-seq datasets for evaluation.

6. Evaluate the performance of added algorithms with provided metrics and datasets.

Difficulty Level: Medium

Prospective contributors should possess or be open to learning the BEELINE framework, gene regulatory networks inference, and single cell multi-omics data analysis.

Size and Length of Project

• small: 90 hours
• 10 weeks

Skills

Essential skills: Python, R, Docker Nice to have skills: Pytorch, Linux, knowledge of single-cell transcriptomics, single-cell chromatin accessibility, and gene regulatory networks

Public Repository

BEELINE https://github.com/Murali-group/Beeline

Potential Mentors

Yiqi Su (yiqisu@vt.edu)

[KarthikDani]

Hello @yiqisu @khanspers!

I find this project quite interesting and different. I'm going to be on my internship at Indian Institute of Science in the Computational and Experimental Biology. I'd like to work on this project but I am worried if I can carry out the work in Mac OS (I've skimmed through the papers and found the requirement of Ubuntu 18)!

Please let me know what do you think. Any suggestions are welcome!

[yiqisu]

Hi @KarthikDani,

Thanks for your interest in this project. There could be a potential issue with the Apple M1 chip, otherwise macOS should be okay. Nonetheless, our objective is to ensure compatibility with various systems.

Please let me know if you have any further questions.

Best, Yiqi

[khanspers]

This is an active GSoC 2024 project. Closing this project idea as it is no longer available to other contributors.