Parameter tuning is a key part of dimensionality reduction via deep variational autoencoders for single cell RNA transcriptomics, bioRxiv, 2018-08-06

Single-cell RNA sequencing (scRNA-seq) is a powerful tool to profile the transcriptomes of a large number of individual cells at a high resolution. These data usually contain measurements of gene expression for many genes in thousands or tens of thousands of cells, though some datasets now reach the million-cell mark. Projecting high-dimensional scRNA-seq data into a low dimensional space aids downstream analysis and data visualization. Many recent preprints accomplish this using variational autoencoders (VAE), generative models that learn underlying structure of data by compress it into a constrained, low dimensional space. The low dimensional spaces generated by VAEs have revealed complex patterns and novel biological signals from large-scale gene expression data and drug response predictions. Here, we evaluate a simple VAE approach for gene expression data, Tybalt, by training and measuring its performance on sets of simulated scRNA-seq data. We find a number of counter-intuitive performance features i.e., deeper neural networks can struggle when datasets contain more observations under some parameter configurations. We show that these methods are highly sensitive to parameter tuning when tuned, the performance of the Tybalt model, which was not optimized for scRNA-seq data, outperforms other popular dimension reduction approaches – PCA, ZIFA, UMAP and t-SNE. On the other hand, without tuning performance can also be remarkably poor on the same data. Our results should discourage authors and reviewers from relying on self-reported performance comparisons to evaluate the relative value of contributions in this area at this time. Instead, we recommend that attempts to compare or benchmark autoencoder methods for scRNA-seq data be performed by disinterested third parties or by methods developers only on unseen benchmark data that are provided to all participants simultaneously because the potential for performance differences due to unequal parameter tuning is so high.

biorxiv bioinformatics 0-100-users 2018

Molecular recording of mammalian embryogenesis, bioRxiv, 2018-08-03

Understanding the emergence of complex multicellular organisms from single totipotent cells, or ontogenesis, represents a foundational question in biology. The study of mammalian development is particularly challenging due to the difficulty of monitoring embryos in utero, the variability of progenitor field sizes, and the indeterminate relationship between the generation of uncommitted progenitors and their progression to subsequent stages. Here, we present a flexible, high information, multi-channel molecular recorder with a single cell (sc) readout and apply it as an evolving lineage tracer to define a mouse cell fate map from fertilization through gastrulation. By combining lineage information with scRNA-seq profiles, we recapitulate canonical developmental relationships between different tissue types and reveal an unexpected transcriptional convergence of endodermal cells from extra-embryonic and embryonic origins, illustrating how lineage information complements scRNA-seq to define cell types. Finally, we apply our cell fate map to estimate the number of embryonic progenitor cells and the degree of asymmetric partitioning within the pluripotent epiblast during specification. Our approach enables massively parallel, high-resolution recording of lineage and other information in mammalian systems to facilitate a quantitative framework for describing developmental processes.

biorxiv developmental-biology 0-100-users 2018

Reaction times and other skewed distributions problems with the mean and the median, bioRxiv, 2018-08-03

ABSTRACTTo summarise skewed (asymmetric) distributions, such as reaction times, typically the mean or the median are used as measures of central tendency. Using the mean might seem surprising, given that it provides a poor measure of central tendency for skewed distributions, whereas the median provides a better indication of the location of the bulk of the observations. However, the sample median is biased with small sample sizes, it tends to overestimate the population median. This is not the case for the mean. Based on this observation, Miller (1988) concluded that “sample medians must not be used to compare reaction times across experimental conditions when there are unequal numbers of trials in the conditions.” Here we replicate and extend Miller (1988), and demonstrate that his conclusion was ill-advised for several reasons. First, the median’s bias can be corrected using a percentile bootstrap bias correction. Second, a careful examination of the sampling distributions reveals that the sample median is median unbiased, whereas the mean is median biased when dealing with skewed distributions. That is, on average the sample mean estimates the population mean, but typically this is not the case. In addition, simulations of false and true positives in various situations show that no method dominates. Crucially, neither the mean nor the median are sufficient or even necessary to compare skewed distributions. Different questions require different methods and it would be unwise to use the mean or the median in all situations. Better tools are available to get a deeper understanding of how distributions differ we illustrate a powerful alternative that relies on quantile estimation. All the code and data to reproduce the figures and analyses in the article are available online.

biorxiv neuroscience 100-200-users 2018

 

Created with the audiences framework by Jedidiah Carlson

Powered by Hugo