Editors Assessment:
This Technical Release (software/workflow paper) presents Poly Pipeline, an open-source spatial transcriptomics workflow designed to handle the data sparsity, fragmented proprietary formats and duplicated-gene complexity that make polyploid plant genomes difficult to analyze. Carvalho et al. combine a converter that turns h5ad and GEM files into the GEF format used by Stereopy, quality filtering, Louvain, Leiden and Spatial Leiden clustering, a user-defined “anchoring” step for assigning regions, and highly variable and marker gene detection. Optional hdWGCNA co-expression networks are built on KNN-derived metacells, intended to retain low-expressed duplicated genes that would otherwise be discarded as dropouts. The pipeline was developed on a hexaploid wheat seed Stereo-seq dataset and benchmarked on rice, Arabidopsis and mouse data in different formats. Spatial Leiden gave more cohesive tissue domains than standard Leiden (15 versus 26 clusters in wheat), and the largest dataset (40,248 cells, 70,766 genes) ran in about 23.5 minutes using a peak of 31.8 GB of RAM. The benchmarking mainly shows that the workflow runs reliably and produces spatially coherent domains across organisms and formats. It does not directly test the claim that hdWGCNA better preserves duplicated-gene signatures, so that remains largely supported by argument rather than comparison with other methods. Nonetheless, with the code available on GitHub under an MIT license (with Zenodo, WorkflowHub, RRID and bio.tools entries), this provides a reproducible, HPC-ready framework that should help plant researchers working with complex genomes and multi-platform spatial datasets.
This evaluation refers to version 1 of the preprint