tauFisher predicts circadian time from a single sample of bulk and single-cell pseudobulk transcriptomic data

Junyan Duan, Michelle N. Ngo, Satya Swaroop Karri, Lam C. Tsoi, Jóhann E. Guðjónsson, Babak Shahbaba, John Lowengrub, Bogi AndersenNature Communications 2024 · DOI

TL;DR

Researchers have developed tauFisher, a computational pipeline that predicts the internal biological (circadian) time of a tissue from just a single transcriptomic sample. Unlike previous methods that require time-series data or multiple samples, tauFisher is platform-agnostic and works across both bulk and single-cell "pseudobulk" data, offering a powerful new lens for personalized circadian medicine.

What Problem Does This Paper Address?

Most organisms have an internal 24-hour clock that regulates everything from metabolism to immune response. However, in patients with chronic diseases or those working night shifts, this internal clock often becomes desynchronized from the external environment.

In clinical research, knowing a sample's "internal time" is vital—it can determine the optimal time to administer a drug or explain why a certain gene is expressed differently in two patients. The challenge? Most existing tools for "timestamping" samples require a full time-series (multiple samples over 24 hours) or are highly sensitive to the specific technology used to sequence the RNA (e.g., they work for microarrays but fail for RNA-seq). tauFisher aims to provide an accurate "timestamp" from a single, isolated sample, regardless of the sequencing platform.

Methods

The tauFisher pipeline consists of a training phase and a testing phase, centered on five key steps:

  1. Feature Selection: Identifying "rhythmic genes" with a reliable 24-hour period.
  2. Functional Data Analysis (FDA): Using curve-fitting to smooth out noise in the training data and fill in missing time points.
  3. Within-Sample Normalization: This is the "secret sauce." Instead of looking at absolute expression levels, tauFisher calculates the difference in expression between pairs of rhythmic genes. This makes the tool platform-agnostic because the relative relationship between genes is more stable than absolute counts.
  4. Dimensionality Reduction: Applying PCA to these gene-pair differences.
  5. Multinomial Regression: Fitting a model to the first two principal components to map the sample onto a 24-hour circular clock.

Datasets

The authors validated tauFisher using a massive array of data:

  • 12 Benchmark Datasets: Covering various tissues (kidney, liver, brainstem, cerebellum, blood, skin, and the SCN) across mice and humans.
  • Cross-Platform Testing: Training on microarray data and testing on RNA-seq data (and vice versa).
  • New scRNA-seq Data: The authors generated a new single-cell RNA-seq dataset from mouse dermal skin collected every 4 hours for 72 hours to test tauFisher’s ability to handle "pseudobulk" (aggregated single-cell) data.

Key Insights and Ideas

  • Platform Agnosticism: tauFisher can be trained on old, abundant microarray data and successfully predict time for modern, expensive scRNA-seq data without retraining. This "transfer learning" capability is rare in transcriptomic tools.
  • The "Dampened" Clock Mystery: When looking at dermal immune cells, the researchers noticed the collective circadian rhythm seemed "dampened" compared to fibroblasts.
  • Phase Heterogeneity: Using a bootstrapping approach with tauFisher, they discovered that the immune cells' clock isn't necessarily "weaker"; rather, individual immune cells are simply less synchronized with each other (higher phase heterogeneity). This suggests that different cell types in the same tissue might be "marching to the beat of different drummers."
  • Robustness to Noise: By focusing on gene-pair correlations rather than absolute values, the model remains accurate even when the input data is "noisy" or unnormalized, which is common in single-cell experiments.

Key Results

  • Superior Accuracy: tauFisher outperformed state-of-the-art methods like ZeitZeiger and TimeSignatR. In 11 out of 12 benchmark datasets, it achieved higher accuracy, typically predicting the time within a 2-hour window of the true label.
  • Single-Cell Success: When applied to mouse SCN (the brain's master clock) and skin pseudobulk data, tauFisher achieved an accuracy of ~78-83%, significantly better than existing tools which often failed to produce a prediction at all due to linear dependency issues.
  • Clinical Relevance: The tool successfully identified circadian disruptions in mice subjected to time-restricted feeding, proving it can distinguish between "healthy" and "disturbed" internal clocks.

Limitations and Future Work

While tauFisher is a leap forward, the authors note a few hurdles:

  • Human Blood Variability: Performance on human blood samples was lower than in other tissues. This is likely due to high inter-individual variability caused by diet, stress, and lifestyle, which "muddies" the transcriptomic clock.
  • Not Truly "Single-Cell": Currently, tauFisher requires "pseudobulk" data (aggregating many cells of the same type). It cannot yet predict the time for a single individual cell because of the high "dropout" rate (missing data) in single-cell sequencing.
  • Future Directions: The authors suggest that incorporating data imputation (predicting missing gene values) could eventually allow tauFisher to work at the true single-cell level, providing even higher resolution into how our biological clocks function.

Summary updated

Abstract

As the circadian clock regulates fundamental biological processes, disrupted clocks are often observed in patients and diseased tissues. Determining the circadian time of the patient or the tissue of focus is essential in circadian medicine and research. Here we present tauFisher, a computational pipeline that accurately predicts circadian time from a single transcriptomic sample by finding correlations between rhythmic genes within the sample. We demonstrate tauFisher's performance in adding timestamps to both bulk and single-cell transcriptomic samples collected from multiple tissue types and experimental settings. Application of tauFisher at a cell-type level in a single-cell RNAseq dataset collected from mouse dermal skin implies that greater circadian phase heterogeneity may explain the dampened rhythm of collective core clock gene expression in dermal immune cells compared to dermal fibroblasts. Given its robustness and generalizability across assay platforms, experimental setups, and tissue types, as well as its potential application in single-cell RNAseq data analysis, tauFisher is a promising tool that facilitates circadian medicine and research.
BibTeX
@inproceedings{duan2024taufisher,
  author = {Junyan Duan and Michelle N. Ngo and Satya Swaroop Karri and Lam C. Tsoi and Jóhann E. Guðjónsson and Babak Shahbaba and John Lowengrub and Bogi Andersen},
  title = {{tauFisher predicts circadian time from a single sample of bulk and single-cell pseudobulk transcriptomic data}},
  booktitle = {Nature Communications},
  year = {2024},
  doi = {10.1038/s41467-024-48041-6},
}
Figure