NNICE – a deep quantile neural network algorithm for expression deconvolution

Understanding the different types of cells in our bodies and their proportions is crucial for assessing health and diagnosing diseases. Recent advancements in gene expression analysis and single-cell RNA sequencing (scRNA-seq) technologies have revolutionized our ability to study cell composition in a detailed and cost-effective manner. Now, researchers at the University of Manitoba have developed a new method called the Neural Network Immune Contexture Estimator (NNICE) that leverages these advancements to estimate cell type abundance from bulk RNA sequencing (RNA-seq) data.

The Importance of Cell Composition

Cells are the building blocks of our bodies, and different types of cells perform various functions. For example, immune cells fight infections, while neurons transmit signals in the brain. Knowing the proportion of different cell types in a tissue sample can provide valuable insights into the state of health or the presence of disease. Traditionally, this has been a challenging and expensive task, but new technologies are changing the game.

Advances in Gene Expression Analysis

Bulk RNA-seq measures the average gene expression levels in a tissue sample, which contains a mix of various cell types. While powerful, bulk RNA-seq does not provide information on individual cell types. In contrast, single-cell RNA sequencing (scRNA-seq) can measure gene expression in individual cells, offering a detailed view of cell composition. However, scRNA-seq is often more expensive and time-consuming compared to bulk RNA-seq.

To bridge this gap, the researchers have developed computational methods to “deconvolve” bulk RNA-seq data, meaning they can estimate the proportions of different cell types from the mixed signals of bulk data. This is where NNICE comes into play.

Introducing NNICE: A New Computational Tool

NNICE, which stands for Neural Network Immune Contexture Estimator, is a novel method developed to estimate cell type abundance from bulk RNA-seq data. NNICE combines quantile regression, a statistical technique, with deep learning, a type of artificial intelligence, to achieve highly accurate and interpretable results.

How NNICE Works

  1. Training the Model: NNICE is trained on a dataset containing both bulk and single-cell RNA-seq data. During training, the model learns the relationship between bulk gene expression profiles and the known cell type proportions from the single-cell data.
  2. Deconvolution Process: Once trained, NNICE can take a bulk RNA-seq sample and estimate the proportions of different cell types within that sample. This process is called deconvolution.
  3. Performance and Accuracy: In tests, NNICE has shown impressive performance, with a Pearson correlation coefficient of 0.9. This means NNICE’s estimates are highly consistent with the actual cell type proportions, both for synthetic (pseudo-bulk) and real bulk gene expression data.

Illustration of Neural Network Immune Contexture Estimator (NNICE)
expression deconvolution model

Figure 1

Different quantiles (10%, 25%, 50%, 75%, 90%) were selected from the bulk gene expression profile. Each of these quantiles was used to train independent ANNs for each cell type of interest (B cells, CD4+ T cells, CD8+ T cells, myeloid cells, NK cells, and other cells). Features from these ANNs were aggregated and input into the final expression deconvolution model for cell type composition prediction.

Why NNICE is a Game-Changer

NNICE stands out for several reasons:

  • Accuracy: It provides highly accurate estimates of cell type proportions.
  • Interpretable Results: The combination of statistical methods with deep learning ensures that the results are not just accurate but also interpretable.
  • Efficiency: NNICE makes it feasible to use more affordable bulk RNA-seq data to gain insights that previously required the more expensive scRNA-seq.

The development of NNICE represents a significant advancement in the field of computational biology. By enabling accurate and interpretable cell type deconvolution from bulk RNA-seq data, NNICE opens up new possibilities for researchers and clinicians. This tool can help in understanding the cellular makeup of tissues in health and disease, potentially leading to better diagnostics and treatments. As we continue to refine these computational methods, the future of personalized medicine and targeted therapies looks increasingly bright.

Jin YW, Hu P, Liu Q. (2024) NNICE: a deep quantile neural network algorithm for expression deconvolution. Sci Rep 14(1):14040. [article]

Leave a Reply

Your email address will not be published. Required fields are marked *


Time limit is exhausted. Please reload CAPTCHA.