A groundbreaking method for analyzing single-cell RNA sequencing (scRNA-seq) data has been introduced by researchers from the National University of Singapore (NUS). This new approach aims to improve the accuracy and speed of interpreting scRNA-seq data, which could significantly enhance research in several fields of biomedicine, particularly in cancer and Alzheimer’s disease studies.
A groundbreaking method for analyzing single-cell RNA sequencing (scRNA-seq) data has been introduced by researchers from the National University of Singapore (NUS). This new approach aims to improve the accuracy and speed of interpreting scRNA-seq data, potentially significantly enhancing research in various fields of biomedicine, particularly in studies related to cancer and Alzheimer’s disease.
The innovative framework, named scAMF (Single-cell Analysis via Manifold Fitting), was created by a research team led by Associate Professor Zhigang Yao from the Department of Statistics and Data Science in the NUS Faculty of Science. This framework applies advanced mathematical methods to fit a low-dimensional manifold in the complex high-dimensional space of gene expression data. By doing this, scAMF effectively reduces noise while preserving vital biological information, leading to more precise identification of cell types and their states.
This study was conducted in partnership with Professor Yau Shing-Tung from Tsinghua University. The results of their work were published in the Proceedings of the National Academy of Sciences on 3 September 2024.
Utilizing manifold fitting techniques to tackle data analysis challenges
Single-cell RNA sequencing is an essential tool in genomic research, providing unparalleled insights into cellular differences and mechanisms of diseases. However, the noise inherent in scRNA-seq data, stemming from biological variability and technical errors, has presented long-standing difficulties for accurate analysis. Traditional scRNA-seq analysis techniques, such as genomic imputation, graph-based methods, and deep learning algorithms, often struggle to delineate cell relationships due to this noise.
The scAMF framework marks a significant advancement in addressing these challenges. It operates on the principle of fitting a low-dimensional manifold within the ambient space of gene expression data, effectively diminishing noise while retaining crucial information. At the core of scAMF is a manifold fitting module that denoises scRNA-seq data by unfolding its distribution in the ambient space. This approach aims to reconstruct a smooth manifold within the original measurement space, capturing the low-dimensional structure in a way that minimizes data loss and effectively mitigates noise.
The primary innovation of scAMF lies in its capacity to enhance the spatial arrangement of the data, drawing together gene expression vectors of similar cell types while ensuring clear distinctions between different cell types. This improvement results in more accurate and reliable clustering during subsequent analyses.
“Our method effectively reduces noise in scRNA-seq data by fitting a low-dimensional manifold within the high-dimensional space,” said Assoc Prof Yao. “This approach significantly boosts the accuracy of cell type classification and improves the clarity of data visualization.”
The scAMF method employs a distinctive blend of data transformation, manifold fitting through shared nearest neighbor metrics, and validation of unsupervised clustering. When compared with other techniques, scAMF showcases outstanding performance in several important areas, such as more efficient noise reduction, enhanced clustering accuracy, superior preservation of biological data, competitive computational efficiency, clearer visual representation, and consistent performance across a variety of datasets. These advancements position scAMF as a formidable tool in single-cell analysis, potentially allowing researchers to uncover previously hidden cellular diversity and rare cell populations.
Future endeavors — Advancing understanding of cellular diversity and functionality
Following the success of scAMF, the research team is now focused on developing a new framework for creating high-resolution, multiscale cell atlases. This updated approach seeks to overcome current challenges in constructing cell atlases, such as difficulties in identifying small cell populations and outdated unsupervised learning methodologies.
A significant emphasis is on creating a multi-resolution cell analysis framework based on scAMF. This advanced framework intends to recognize rare cell populations and facilitate the development of detailed cell atlases. The multi-resolution strategy will allow researchers to examine cellular diversity at various depth levels, from broad cell categories to subtle subpopulations. This capability is critical for detecting rare cell types that might be missed with conventional analytical techniques.
“Our ongoing research has already yielded promising outcomes across numerous benchmark datasets, leading to new biological discoveries,” noted Assoc Prof Yao. “We have applied it to the Human Brain Cell Atlas, discovering new subtypes and marker genes for various cell types.”
This continuing research holds the potential to further expand the horizons of single-cell analysis, potentially transforming our comprehension of cellular diversity and operations across numerous biological systems.
