iRECODE Reduces Even More Noise in Single-Cell Research

Studying cells at high resolution has revealed new layers of biological complexity, but analysis is often hindered by noise that obscures signals. Each cell carries unique genetic information, yet technical and experimental variation can blur results and make rare cell types or subtle early disease markers difficult to detect.

Single-cell RNA sequencing is a widely used tool to assess gene expression at the individual cell level, but its accuracy is limited by two main sources of noise. Technical noise arises from measurement gaps, such as the “dropout effect,” in which some expressed genes go undetected. Batch noise comes from differences between experimental conditions or instruments, introducing inconsistencies across datasets. Both factors make it difficult to capture the true state of cells.

In 2022, Yusuke Imoto and colleagues at Kyoto University’s Institute for the Advanced Study of Human Biology developed RECODE (resolution of the curse of dimensionality) to reduce technical noise in these high-dimensional datasets. Because single-cell experiments involve thousands of genes measured per cell, random fluctuations can overwhelm true biological signals. RECODE applies statistical methods tailored to this challenge, refining gene expression data so it more closely reflects expected values without requiring intricate parameter settings or machine learning.

Expanding on this work, Prof. Imoto recently created iRECODE (Integrative RECODE), which simultaneously reduces both technical and batch noise. Applied to single-cell RNA sequencing, iRECODE refined gene expression distributions, resolved the sparsity caused by missing values, and improved the consistency of cell types across batches while maintaining their distinct identities. Tests showed that iRECODE performed about ten times more efficiently than applying noise reduction and batch correction separately.

The method, described in Cell Reports Methods, also extends across single-cell technologies, including Drop-seq, Smart-Seq, and 10x Genomics protocols. In spatial transcriptomics, iRECODE clarified noisy data to reveal tissue-level patterns of cellular interaction. In scHi-C datasets, which map chromosomal interactions but often suffer from sparse data, RECODE reduced noise and uncovered real contacts, aiding downstream clustering when paired with machine learning approaches.

By effectively lowering noise across a range of data types, iRECODE offers researchers a clearer view of cellular behavior. “Single-cell data captures countless cellular ‘whispers,’ but hearing those whispers through the noise is extremely difficult,” said Imoto. “iRECODE, an evolution of our RECODE method, lifts those voices to the surface. Through this method, I believe the hidden stories of cells, stories we could never hear before, will steadily come to light."