lute, a new framework for bulk transcriptomics deconvolution experiments

Author(s): Sean Maden,Stephanie Hicks

Affiliation(s): Johns Hopkins Bloomberg School of Public Health

Social media: https://twitter.com/MadenSean

Each year sees the publication of many new and innovative algorithms to deconvolve cell type amounts from bulk tissues using single-cell RNA-seq reference datasets. Many of these new algorithms are published to GitHub and Bioconductor, have several of their own dependencies, and are written in different programming languages. Further, standard checks for deconvolution experiments are commonly rewritten, such as cell type summaries and checks for the bulk signals matrix, signature matrix, and cell size factors. To make it easier to perform bulk deconvolution using single-cell RNA-seq references, we introduced the lute R package. Using a similar approach to the bluster package for clustering algorithms, we defined a new deconvolution generic function, a series of classes for each algorithm, and methods for each class to map each standard input to its specific algorithm synonym. In this way, lute can help with greater accessibility, understandability, and comparability of bulk deconvolution algorithms. lute uses a hierarchical class-and-method structure to limit the need to recode common experiment tasks. It further provides a means of always accounting for cell size bias prior to running a given deconvolution algorithm, an important step that is being explored in new bulk transcriptomics methods. These measures should facilitate the support of more deconvolution algorithms and the development of altogether new deconvolution algorithms. lute further provides new utilities for benchmarks of R functions with NextFlow workflows, conducting new simulations using either simulated cell types or batch-effect biases, and reliably running algorithm-specific conda environments. While lute is still under active development, we anticipate it will facilitate more standardized method benchmarks and straightforward algorithm fine-tuning in bulk deconvolution experiments. We wish to present lute in a Short Talk at Bioc2023, to introduce the package to the Bioconductor community and obtain feedback on how we can further improve its utility.