MOTIVATION: Differential expression and marker gene selection methods for single-cell RNA-sequencing (scRNA-seq) data can struggle to identify small sets of informative genes, especially for subtle differences between cell states, as can be induced by disease or treatment. RESULTS: We present iterative logistic regression (iLR) for the identification of small sets of informative marker genes. iLR applied logistic regression iteratively with a Pareto front optimization to balance gene set size with classification performance. Benchmarking iLR on in silico datasets, we demonstrated its comparable performance to the state-of-the-art at single-cell classification using only a fraction of the genes. We then tested iLR on its ability to distinguish neuronal cell subtypes in healthy versus autism spectrum disorder patients and find that it achieves high accuracy with small sets of disease-relevant genes. Applying iLR to investigate immunotherapeutic effects in cell types from different tumor microenvironments, we found that iLR infers informative genes that translate across organs and even species (mouse-to-human) comparison. We predicted via iLR that entinostat acts in part through the modulation of myeloid cell differentiation routes in the lung microenvironment. Overall, iLR provides means to infer interpretable transcriptional signatures from complex datasets with prognostic or therapeutic potential. AVAILABILITY AND IMPLEMENTATION: iLR is freely available at GitHub https://github.com/maclean-lab/iLR and Zenodo https://zenodo.org/records/17728797.

Biological systems exhibit complex dynamics that differential equations can often adeptly represent. Ordinary differential equation models are widespread; until recently their construction has required extensive prior knowledge of the system. Machine learning methods offer alternative means of model construction: differential equation models can be learnt from data via model discovery using sparse identification of nonlinear dynamics (SINDy). However, SINDy struggles with realistic levels of biological noise and is limited in its ability to incorporate prior knowledge of the system. We propose a data-driven framework for model discovery and model selection using hybrid dynamical systems: partial models containing missing terms. Neural networks are used to approximate the unknown dynamics of a system, enabling the denoising of the data while simultaneously learning the latent dynamics. Simulations from the fitted neural network are then used to infer models using sparse regression. We show, via model selection, that model discovery using hybrid dynamical systems outperforms alternative approaches. We find it possible to infer models correctly up to high levels of biological noise of different types. We demonstrate the potential to learn models from sparse, noisy data in application to a canonical cell state transition using data derived from single-cell transcriptomics. Overall, this approach provides a practical framework for model discovery in biology in cases where data are noisy and sparse, of particular utility when the underlying biological mechanisms are partially but incompletely known.

MOTIVATION: Methods to model dynamic changes in gene expression at a genome-wide level are not currently sufficient for large (temporally rich or single-cell) datasets. Variational autoencoders offer means to characterize large datasets and have been used effectively to characterize features of single-cell datasets. Here, we extend these methods for use with gene expression time series data. RESULTS: We present RVAgene: a recurrent variational autoencoder to model gene expression dynamics. RVAgene learns to accurately and efficiently reconstruct temporal gene profiles. It also learns a low dimensional representation of the data via a recurrent encoder network that can be used for biological feature discovery, and from which we can generate new gene expression data by sampling the latent space. We test RVAgene on simulated and real biological datasets, including embryonic stem cell differentiation and kidney injury response dynamics. In all cases, RVAgene accurately reconstructed complex gene expression temporal profiles. Via cross validation, we show that a low-error latent space representation can be learnt using only a fraction of the data. Through clustering and gene ontology term enrichment analysis on the latent space, we demonstrate the potential of RVAgene for unsupervised discovery. In particular, RVAgene identifies new programs of shared gene regulation of Lox family genes in response to kidney injury. AVAILABILITY AND IMPLEMENTATION: All datasets analyzed in this manuscript are publicly available and have been published previously. RVAgene is available in Python, at GitHub: https://github.com/maclean-lab/RVAgene; Zenodo archive: http://doi.org/10.5281/zenodo.4271097. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Motivation: Single-cell RNA-sequencing (scRNA-seq) offers unprecedented resolution for studying cellular decision-making processes. Robust inference of cell state transition paths and probabilities is an important yet challenging step in the analysis of these data. Results: Here we present scEpath, an algorithm that calculates energy landscapes and probabilistic directed graphs in order to reconstruct developmental trajectories. We quantify the energy landscape using 'single-cell energy' and distance-based measures, and find that the combination of these enables robust inference of the transition probabilities and lineage relationships between cell states. We also identify marker genes and gene expression patterns associated with cell state transitions. Our approach produces pseudotemporal orderings that are-in combination-more robust and accurate than current methods, and offers higher resolution dynamics of the cell state transitions, leading to new insight into key transition events during differentiation and development. Moreover, scEpath is robust to variation in the size of the input gene set, and is broadly unsupervised, requiring few parameters to be set by the user. Applications of scEpath led to the identification of a cell-cell communication network implicated in early human embryo development, and novel transcription factors important for myoblast differentiation. scEpath allows us to identify common and specific temporal dynamics and transcriptional factor programs along branched lineages, as well as the transition probabilities that control cell fates. Availability and implementation: A MATLAB package of scEpath is available at https://github.com/sqjin/scEpath. Supplementary information: Supplementary data are available at Bioinformatics online.