Single-cell biological network inference using a heterogeneous graph transformer

Single-cell biological network inference using a heterogeneous graph transformer
Like

Artificial intelligence (AI) and single-cell studies have been making waves in the science and technology communities. By sequencing the genetic material of one cell at a time, researchers can uncover the unique features and characteristics that make each cell in our body special. This single-cell sequencing technique has already revolutionized our understanding of the brain, cancer, and the immune system. AI offers a broad range of methods that can be used to investigate diverse data- and hypothesis-driven questions in single-cell biology. Specifically, the highly heterogeneous nature of single-cell sequencing data can be analyzed across a wide range of research topics by generalizing deep-learning model design and optimization. However, there is a catch. Each type of single-cell sequencing data only gives us a glimpse of what is happening inside a cell. It is like looking at a tiny piece of a puzzle - we can see some details but do not get the whole picture. That is where a new approach called single-cell multi-omics comes in. By analyzing multiple types of data from the same cell, researchers can better understand how cells work and what makes them unique.

 

Several tools have been developed to help scientists make sense of the vast amounts of data generated by single-cell sequencing, e.g., Seurat, MOFA, and Harmony. These tools can predict what types of cells are present in a sample, remove unwanted variations between different batches of cells, and identify connections between different types of data. However, most of these tools do not consider the complex relationships between cells and the different types of data (e.g., genes, proteins, and enhancers) being analyzed. Graph neural network (GNN) is a type of deep learning algorithm good for modeling and analyzing single-cell sequencing data with a graph representation learning style. For example, this technique allows them to build a graph that connects different cells and genes based on their similarity. It can also identify the active biological networks that control how cells function.

 

DeepMAPS (Deep learning-based Multi-omics Analysis Platform for Single-cell data) is newly developed to help scientists analyze and interpret single-cell multi-omics data. It uses a cutting-edge GNN model, specifically the heterogeneous graph transformer (HGT), to model single-cell multi-omics data and identify the active biological networks that control cells' functions. Specifically, it creates a graph that includes both cells and genes as nodes, with their relationships acting as edges. The HGT model looks at all the connections between cells and genes to understand how they interact, without making any assumptions about which genes work together. One cool thing about HGT is that it can estimate which genes are most important for specific cells, which helps scientists understand how cells work and the underlying molecular program. Specifically, DeepMAPS outperforms other methods in terms of accuracy and efficiency, and it is effective at analyzing data from lung tumors and other types of cancers. To make DeepMAPS more accessible, we have also created a web server that provides a range of useful features and visualizations. With this tool, scientists can easily explore their data and gain new insights into the complex world of cellular biology. In addition, DeepMAPS can infer two biological networks: gene association and regulatory networks (GRN). To measure the importance of genes for cell function, we use "centrality scores" and "functional enrichment." In one study, we compared the gene association networks created by DeepMAPS to those created by other tools and found that the networks created by DeepMAPS were more closely connected and more relevant to cell function. Our research indicates how DeepMAPS can help us learn more about the genes and networks that make up our cells, and how they work together to keep us healthy.

 

Our team at the Bioinformatics and Mathematical Biosciences Lab (BMBL) of the Ohio State University focus on the research of single-cell multi-omics data and are very excited to deliver and further maintain DeepMAPS. Users can find the DeepMAPS source code at: https://github.com/OSU-BMBL/deepmaps. The web server is available at: https://bmblx.bmi.osumc.edu/. To read more about our approach and methods, check the DeepMAPS paper published on Nature Communications: https://www.nature.com/articles/s41467-023-36559-0. As a long-term goal, we aim to develop cutting-edge computational tools to discover underlying molecular mechanisms in diverse biological systems and complex diseases. For more information, visit https://u.osu.edu/bmbl/.

DeepMAPS overview
DeepMAPS is a Deep learning-based Multi-omics Analysis Platform for Single-cell data. It allows for the joint analysis of multiple scRNA-seq, CITE-seq (matched RNA and protein profiling), and matched single-cell RNA and ATAC-seq (scRNA-ATAC-seq) datasets. The core deep learning method includes the representation of cell-gene relations via a heterogeneous graph and a transformer with a graph attention mechanism. DeepMAPS provides interactive and interpretable graphical representations to deliver cell clusters and cell-type-specific biological networks based on modality types. DeepMAPS also provides a web portal to ensure robustness and reproducibility, along with a docker container. Workspace is provided for job saving and retrieval.

Please sign in or register for FREE

If you are a registered user on Research Communities by Springer Nature, please sign in

Subscribe to the Topic

Biological Techniques
Life Sciences > Biological Sciences > Biological Techniques

Related Collections

With collections, you can get published faster and increase your visibility.

Pre-clinical drug discovery

We welcome studies reporting advances in the discovery, characterization and application of compounds active on biologically or industrially relevant targets. Examples include emerging screening technologies, the development of small bioactive compounds/peptides/proteins, and the elucidation of compound structure-activity relationships, target interactions and mechanism-of-action.

Publishing Model: Open Access

Deadline: Mar 31, 2024

Biomedical applications for nanotechnologies

Overall, there are still several challenges on the path to the clinical translation of nanomedicines, and we aim to bridge this gap by inviting submissions of articles that demonstrate the translational potential of nanomedicines with promising pre-clinical data.

Publishing Model: Open Access

Deadline: Mar 31, 2024