Representation learning of RNA velocity reveals robust cell transitions

RNA velocity is a promising technique for quantifying cellular transitions from single-cell transcriptome experiments and revealing transient cellular dynamics among a heterogeneous cell population. However, the cell transitions estimated from high-dimensional RNA velocity are often unstable or inaccurate, partly due to the high technical noise and less informative projection. Here, we present Velocity Autoencoder (VeloAE), a tailored representation learning method, to learn a low-dimensional representation of RNA velocity on which cellular transitions can be robustly estimated. On various experimental datasets, we show that VeloAE can both accurately identify stimulation dynamics in time-series designs and effectively capture expected cellular differentiation in different biological systems. VeloAE, therefore, enhances the usefulness of RNA velocity for studying a wide range of biological processes.

Building the mega single-cell transcriptome ocular meta-atlas

Background: The development of highly scalable single-cell transcriptome technology has resulted in the creation of thousands of datasets, >30 in the retina alone. Analyzing the transcriptomes between different projects is highly desirable because this would allow for better assessment of which biological effects are consistent across independent studies. However it is difficult to compare and contrast data across different projects because there are substantial batch effects from computational processing, single-cell technology utilized, and the natural biological variation. While many single-cell transcriptome-specific batch correction methods purport to remove the technical noise, it is difficult to ascertain which method functions best. Results: We developed a lightweight R package (scPOP, single-cell Pick Optimal Parameters) that brings in batch integration methods and uses a simple heuristic to balance batch merging and cell type/cluster purity. We use this package along with a Snakefile-based workflow system to demonstrate how to optimally merge 766,615 cells from 33 retina datsets and 3 species to create a massive ocular single-cell transcriptome meta-atlas. Conclusions: This provides a model for how to efficiently create meta-atlases for tissues and cells of interest.

Noise perception assessment in urban green spaces using soundwalk approach

Urban green spaces are intended to provide citizens with calm environments free of annoying city noises. This requires a thorough understanding of noise emission and related exposure to sounds in green spaces. This research investigates noise perception in various spots in an urban green space. For this purpose, the study has been conducted in the grand garden of the city of Dresden. The garden covers 1.8 square kilometers of various landscapes, including water streams, park railways, fountains, bridges, roads for bicycles and pedestrians etc. Noise perception was investigated at eleven spots with emphasis on four noise types: nature noise, human noise, traffic noise, and technical noise. In parallel, audio-visual recordings were conducted for each spot to identify the connection between the perceptual measures and the psychoacoustic parameters. These spots are categorized based on the resulting perception and psychoacoustic parameters. In addition, the visual effect of each spot on final perception is investigated. Eventually, annoyance for each spot is identified based on the corresponding participants' perception and is associated with the relevant psychoacoustic parameters.

Self-supervised contrastive learning for integrative single cell RNA-seq data analysis

Single-cell RNA-sequencing (scRNA-seq) has become a powerful tool to reveal the complex biological diversity and heterogeneity among cell populations. However, the technical noise and bias of the technology still have negative impacts on the downstream analysis. Here, we present a self-supervised Contrastive LEArning framework for scRNA-seq (CLEAR) profile representation and the downstream analysis. CLEAR overcomes the heterogeneity of the experimental data with a specifically designed representation learning task and thus can handle batch effects and dropout events. In the task, the deep learning model learns to pull together the representations of similar cells while pushing apart distinct cells, without manual labeling. It achieves superior performance on a broad range of fundamental tasks, including clustering, visualization, dropout correction, batch effect removal, and pseudo-time inference. The proposed method successfully identifies and illustrates inflammatory-related mechanisms in a COVID-19 disease study with 43,695 single cells from peripheral blood mononuclear cells. Further experiments to process a million-scale single-cell dataset demonstrate the scalability of CLEAR. This scalable method generates effective scRNA-seq data representation while eliminating technical noise, and it will serve as a general computational framework for single-cell data analysis.

Dimensionality reduction and data integration for scRNA-seq data based on integrative hierarchical Poisson factorisation

Single-cell RNA sequencing (scRNA-seq) data sets consist of high-dimensional, sparse and noisy feature vectors, and pose a challenge for classic methods for dimensionality reduction. We show that application of Hierarchical Poisson Factorisation (HPF) to scRNA-seq data produces robust factors, and outperforms other popular methods. To account for batch variability in composite data sets, we introduce Integrative Hierarchical Poisson Factorisation (IHPF), an extension of HPF that makes use of a noise ratio hyper-parameter to tune the variability attributed to technical (batches) vs. biological (cell phenotypes) sources. We exemplify the advantageous application of IHPF under data integration scenarios with varying alignments of technical noise and cell diversity, and show that IHPF produces latent factors with a dual block structure in both cell and gene spaces for enhanced biological interpretability.

Deep learning algorithms for scRNAseq analysis have yielded positive results, but there are still more promising ways that need to be developed for regenerative medicine

A detailed survey of the use of many deep learning algorithms on scRNAseq data for regenerative medicine was published in this article. Currently, the best deep learning algorithms for scRNAseq analysis have yielded positive results, but there are still more promising ways that need to be developed to better handle technical noise, account for cell expression variability, identify MSCs, and anticipate stem cell type. To gain access to scRNAseq data, these deep learning techniques need to be paired with scRNAseq data. The ability to identify cell types and functions accurately and fast utilizing these algorithms has not yet been made possible. In the study we conducted, we reached the conclusion that further research has to be done into how to apply deep learning algorithms to interpret scRNAseq data, which may be used to better cell therapy and regenerative medicine efforts.

Replicate sequencing libraries are important for quantification of allelic imbalance

A sensitive approach to quantitative analysis of transcriptional regulation in diploid organisms is analysis of allelic imbalance (AI) in RNA sequencing (RNA-seq) data. A near-universal practice in such studies is to prepare and sequence only one library per RNA sample. We present theoretical and experimental evidence that data from a single RNA-seq library is insufficient for reliable quantification of the contribution of technical noise to the observed AI signal; consequently, reliance on one-replicate experimental design can lead to unaccounted-for variation in error rates in allele-specific analysis. We develop a computational approach, Qllelic, that accurately accounts for technical noise by making use of replicate RNA-seq libraries. Testing on new and existing datasets shows that application of Qllelic greatly decreases false positive rate in allele-specific analysis while conserving appropriate signal, and thus greatly improves reproducibility of AI estimates. We explore sources of technical overdispersion in observed AI signal and conclude by discussing design of RNA-seq studies addressing two biologically important questions: quantification of transcriptome-wide AI in one sample, and differential analysis of allele-specific expression between samples.

Paired-end small RNA sequencing reveals a possible overestimation in the isomiR sequence repertoire previously reported from conventional single read data analysis

Background Next generation sequencing has allowed the discovery of miRNA isoforms, termed isomiRs. Some isomiRs are derived from imprecise processing of pre-miRNA precursors, leading to length variants. Additional variability is introduced by non-templated addition of bases at the ends or editing of internal bases, resulting in base differences relative to the template DNA sequence. We hypothesized that some component of the isomiR variation reported so far could be due to systematic technical noise and not real. Results We have developed the XICRA pipeline to analyze small RNA sequencing data at the isomiR level. We exploited its ability to use single or merged reads to compare isomiR results derived from paired-end (PE) reads with those from single reads (SR) to address whether detectable sequence differences relative to canonical miRNAs found in isomiRs are true biological variations or the result of errors in sequencing. We have detected non-negligible systematic differences between SR and PE data which primarily affect putative internally edited isomiRs, and at a much smaller frequency terminal length changing isomiRs. This is relevant for the identification of true isomiRs in small RNA sequencing datasets. Conclusions We conclude that potential artifacts derived from sequencing errors and/or data processing could result in an overestimation of abundance and diversity of miRNA isoforms. Efforts in annotating the isomiRnome should take this into account.

A computational solution for bolstering reliability of epigenetic clocks: Implications for clinical trials and longitudinal tracking

Epigenetic clocks are widely used aging biomarkers calculated from DNA methylation data. Unfortunately, measurements for individual CpGs can be surprisingly unreliable due to technical noise, and this may limit the utility of epigenetic clocks. We report that noise produces deviations up to 3 to 9 years between technical replicates for six major epigenetic clocks. The elimination of low-reliability CpGs does not ameliorate this issue. Here, we present a novel computational multi-step solution to address this noise, involving performing principal component analysis on the CpG-level data followed by biological age prediction using principal components as input. This method extracts shared systematic variation in DNAm while minimizing random noise from individual CpGs. Our novel principal-component versions of six clocks show agreement between most technical replicates within 0 to 1.5 years, equivalent or improved prediction of outcomes, and more stable trajectories in longitudinal studies and cell culture. This method entails only one additional step compared to traditional clocks, does not require prior knowledge of CpG reliabilities, and can improve the reliability of any existing or future epigenetic biomarker. The high reliability of principal component-based epigenetic clocks will make them particularly useful for applications in personalized medicine and clinical trials evaluating novel aging interventions.

Theoretical and Experimental Investigation of the Effect of Pump Laser Frequency Fluctuations on Signal-to-Noise Ratio of Brillouin Dynamic Grating Measurement with Coherent FMCW Reflectometry

Signal-dependent speckle-like noise has constituted a serious factor in Brillouin-grating based frequency-modulated continuous-wave (FMCW) reflectometry and it has been indispensable for improving the signal-to-noise ratio (S/N) of the Brillouin dynamic grating measurement to clarify the noise generation mechanism. In this paper we show theoretically and experimentally that the noise is generated by the frequency fluctuations of the pump light from a laser diode (LD). We could increase the S/N from 36 to 190 merely by driving the LD using a current source with reduced technical noise. On the basis of our experimental result, we derived the theoretical formula for S/N as a function of distance, which contained the second and fourth-order moments of the frequency fluctuations, by assuming that the pump light frequency was modulated by the technical noise. We calculated S/N along the 1.35 m long optical fiber numerically using the measured power spectral density of the frequency fluctuations, and the resulting distributions agreed with the measured values in the 10 to 190 range. Since higher performance levels are required if the pump light source is to maintain the S/N as the fiber length increases, we can use the formula to calculate the light source specifications including the spectral width and rms value of the frequency fluctuations to achieve a high S/N while testing a fiber of a given length.