Neural networks could replace marker genes in RNA sequencing

Researchers at Carnegie Mellon University have found a new and efficient way to characterize cell types following single-cell RNA sequencing (scRNA-seq).

ktsdesign | Shutterstock

The new method involves the use of neural networks and supervised machine learning techniques rather than marker genes, which are not available for all cell types.

Using this new, automated technique, researchers can analyze all of the data they need. This provides researchers with the ability to analyze and compare all types of cells.

The authors of the study also describe a web server called scQuery that enables the technique to be used by any researcher.

Over recent years, single-cell sequencing has become a popular tool, enabling researchers to identify subtypes of cells or to see the difference between a healthy and diseased cell or a young versus aged cell.

Previously, scientists could only process batches of cells to obtain results that reflect an overall average of their value.

The new method, which has been described in the newspaper Nature Communicationswill be employed as part of the National Institutes of Health's new Human BioMolecular Atlas Program, which is creating a 3D map of the human body that will show how tissues differ on a cellular level.

Computational biologist Amir Alavi says that the data is only created by a "big data" problem that is not able to manage.

Alavi and colleagues developed an automated pipeline, with the aim of downloading all public scRNA-seq data available on the largest repositories so that the genes and proteins expressed in each cell could be identified.

The cells were then labeled according to a different neural network and modeled on the human brain.

To test the model, Alavi and team used scRNA-seq data from an animal study of a disease similar to Alzheimer's. As anticipated, similar numbers of brain cells have been seen in both healthy and diseased tissues, with

The researchers used their automated pipeline and methods to develop scQuery web server, which accelerates the comparative analysis of new scRNA-seq data.

Once a single-cell experiment is entered into scQuery, the team's neural network and matching methods rapidly identify related subtypes of cells, as well as previous studies of similar cells.