FiRE away and spot the odd cell out

Our body is made up of trillions of cells of all shapes, sizes and kinds. Among this vibrant diversity lie a few outliers, like the circulating tumour cells, cancer stem cells and cells belonging to our immune system. These cells are few and far between, and identifying them among the trillions is a Herculean task.

Although sparse, these rare cells are essential as some play critical roles in our immune responses, replace damaged cells, help treat diseases like Parkinson’s, diabetes and heart diseases and act as potential indicators of other conditions, including cancer. Hence, identifying them becomes important to detect and treat the related ailments.

In a recent study, researchers from the Indian Institute of Technology Delhi, and the Indraprastha Institute of Information Technology, Delhi, have developed an algorithm to identify such rare cells based on their genes. They propose a computer-based algorithm called Finder of Rare Entities (FiRE) which identifies these rare cells in a matter of seconds. The study was published in Nature Communications and was partially funded by the Department of Science and Technology, Government of India.

Molecular biologists use a technique called single-cell RNA sequencing (scRNA-seq) to find out rare cell types within a mass of cells.

Here, various types of cells can be isolated from a small sample, like a single drop of blood, and the pattern of the ‘gene expression profile’, or the measurement of the activity of thousands of genes in each of these cells, is analysed in parallel. The rare cells are identified based on the differences in their gene expression profiles as each type of cell has a unique expression profile.

Remember those days when you were asked to ‘spot the difference’ between two almost-identical images, or pick an ‘odd one out’ from a set of similar images? Well, FiRE does something similar, albeit dealing with thousands of identical things instead of a few.

It looks at all the profiles of all cells in the dataset provided and assigns a ‘rareness score’ to each of them. The rarer the type of a cell, the higher is this score. It then
shortlists the ‘rare’ ones based on their score.

“FiRE assigns a continuous score to each cell, such that outlier cells and cells originating from the minor cell populations are assigned higher values in comparison with cells representing major subpopulations. A continuous score gives users the freedom to decide the degree of the rareness of the cells, to be further investigated,” explain researchers, talking about the novelty of their algorithm. Although there are a couple of well-known algorithms that identify rare cells by grouping those similar in their expression profiles, they are inefficient and do not scale up when applied on large datasets.

Performance evaluation

The researchers evaluated the performance of FiRE using various datasets. The algorithm successfully identified rare cells from a test dataset containing the expression profile of 68,000 cells. “FiRE took around 31 seconds to analyse a single-cell mRNA sequencing dataset containing about 68,000 expression profiles. Such unrivalled speed, combined with the ability to pinpoint the truly rare expression profiles, makes the algorithm future-proof,” say the researchers.

Interestingly, using FiRE, the researchers also identified previously unknown, rare cell subtypes during an experiment with a dataset containing about 20,000 cells from the mouse brain. One subtype of these newly discovered rare cells plays an essential role in the development of the pituitary gland in the brain of mammals, say the researchers.

The new algorithm bears great potential in detecting rare cells and diseases. With the growth in technologies to generate a massive amount of biological data, like the expression profile of genes from individual cells, there is a growing need for the development of tools to analyse and retrieve information in the datasets. Algorithms like FiRE, which are fast and efficient, would greatly benefit the research community.

(The author is with Gubbi Labs, Bengaluru)