MotivationDeep learning techniques have yielded tremendous progress in the field of computational biology over the last decade, however many of these techniques are opaque to the user. To provide interpretable results, methods have incorporated biological priors directly into the learning task; one such biological prior is pathway structure. While pathways represent most biological processes in the cell, the high level of correlation and hierarchical structure make it complicated to determine an appropriate computational representation.ResultsHere, we present pathway module Variational Autoencoder (pmVAE). Our method encodes pathway information by restricting the structure of our VAE to mirror gene-pathway memberships. Its architecture is composed of a set of subnetworks, which we refer to as pathway modules. The subnetworks learn interpretable latent representations by factorizing the latent space according to pathway gene sets. We directly address correlation between pathways by balancing a module-specific local loss and a global reconstruction loss. Furthermore, since many pathways are by nature hierarchical and therefore the product of multiple downstream signals, we model each pathway as a multidimensional vector. Due to their factorization over pathways, the representations allow for easy and interpretable analysis of multiple downstream effects, such as cell type and biological stimulus, within the contexts of each pathway. We compare pmVAE against two other state-of-the-art methods on two single-cell RNA-seq case-control data sets, demonstrating that our pathway representations are both more discriminative and consistent in detecting pathways targeted by a perturbation.Availability and implementationhttps://github.com/ratschlab/pmvae

Unsupervised trajectory inference from single cell RNA sequencing data bears the potential to systematically reconstruct complex differentiation processes but remains challenging notwithstanding the many available solutions. In general, trajectory and pseudotime inference methods have so far suffered from the ambiguity of the static single-cell transcriptome snapshots lacking a measure of directionality of transcriptional activity. We report Cytopath, a method for trajectory inference that takes advantage of transcriptional activity information from RNA velocity of single cells to perform trajectory inference. Cytopath performs this task by defining a Markov chain model, simulation of an ensemble of possible differentiation trajectories and subsequent statistical analysis to extract the topological and molecular characteristics of the studied differentiation process. We demonstrate the capability of Cytopath to reconstruct differentiation trajectories with varying bifurcated and circular topology studied in single-snapshot as well as time-series single-cell RNA sequencing experiments. Comparison to state-of-the art trajectory inference approaches demonstrate superior and enabling capability to reconstruct the considered differentiation trajectories. Trajectory inference constitutes a frequent step in interpreting single-cell RNA sequencing studies. We expect Cytopath to enable researchers to tap the directionality information present in single-cell RNA sequencing data to achieve trajectory inference for possibly complex lineages at an unprecedented precision and resolution.

No abstract