Gene selection for optimal prediction of cell position in tissues from single-cell transcriptomics data

Abstract

Single-cell RNA-sequencing (scRNAseq) technologies are rapidly evolving. Although very informative, in standard scRNAseq experiments, the spatial organization of the cells in the tissue of origin is lost. Conversely, spatial RNA-seq technologies designed to maintain cell localization have limited throughput and gene coverage. Mapping scRNAseq to genes with spatial information increases coverage while providing spatial location. However, methods to perform such mapping have not yet been benchmarked. To fill this gap, we organized the DREAM Single-Cell Transcriptomics challenge focused on the spatial reconstruction of cells from the Drosophila embryo from scRNAseq data, leveraging as silver standard, genes with in situ hybridization data from the Berkeley Drosophila Transcription Network Project reference atlas. The 34 participating teams used diverse algorithms for gene selection and location prediction, while being able to correctly localize clusters of cells. Selection of predictor genes was essential for this task. Predictor genes showed a relatively high expression entropy, high spatial clustering and included prominent developmental genes such as gap and pair-rule genes and tissue markers. Application of the top 10 methods to a zebra fish embryo dataset yielded similar performance and statistical properties of the selected genes than in the Drosophila data. This suggests that methods developed in this challenge are able to extract generalizable properties of genes that are useful to accurately reconstruct the spatial arrangement of cells in tissues.

Machine Learning Approaches Identify Genes Containing Spatial Information From Single-Cell Transcriptomics Data

“…Based on this, our final submission used results based on NN for subchallenge 2 and Lasso.TopX for the other two. Our submitted results ranked 10th, 6th, and 4th in the three subchallenges, respectively, among ∼40 participating teams ( Tanevski et al, 2020 ).…”

“…After the challenge ended, the organizers devised a postchallenge CV scheme [see section “Methods” and Tanevski et al (2020) for more detail] to evaluate the robustness of the methods. It was only after this resubmission phase did the organizers make the true scoring functions (“s1,” “s2,” and “s3” scores) publicly available.…”

“…In the postchallenge phase, the organizers split the data in 10 folds, on which our approaches were rerun for stability and overfitting evaluation ( Tanevski et al, 2020 ). Separately for each of the 10 iterations, only the respective nine training folds were used for feature selection and to train DistMap.…”

“…For more than a decade, the Dialogue on Reverse Engineering Assessment and Methods (DREAM) ( Stolovitzky et al, 2007 ) initiative has driven crowd-sourced open science scientific contests in different areas of biology and medicine. Recently, the DREAM Single-Cell Transcriptomics Challenge ( Tanevski et al, 2020 ), in which we participated, focused on tackling the reconstruction of the 3-D arrangement of cells using predefined number of genes. Specifically, the goal of this DREAM challenge was to use Drosophila melanogaster embryo as a model system and seek to determine whether one can reconstruct the spatial arrangement of cells from a stage 6 embryo by using only a limited number of genes.…”

“…Of these 84 genes (herein referred to as “ inSitu genes”) and without using hybridization data, the participants were asked to identify the most informative 60, 40, and 20 genes for subchallenges #1, #2, and #3, respectively. In addition to gene selection, each subchallenge also required participants to submit 10 locations predictions (X, Y, Z coordinates) for each of the cells using only the selected genes ( Tanevski et al, 2020 ).…”