Handling biological replicates in long-read RNA sequencing data by joining or not joining

While isoform identification from long-read RNA sequencing (lrRNA-seq) data has received significant attention, the handling of biologically replicated lrRNA-seq datasets remains less explored. This study defines two strategies for obtaining consensus transcriptomes from multi-sample lrRNA-seq data: Join & Call, where reads from all samples are combined before transcript reconstruction, and Call & Join, where transcript reconstruction is performed on individual samples before combining the resulting annotations. We apply these strategies to mouse brain and kidney tissue datasets, using PacBio and ONT technologies, across six transcript reconstruction tools. Our results indicate that the optimal strategy depends on the tool and research objective. We find that Join & Call is generally preferable for discovering novel isoforms, while Call & Join is often preferable for highly replicated datasets when the discovery of novelty is secondary. Our findings provide a conceptual and practical framework for multi-sample transcript reconstruction, guiding best practices for increasingly large-scale lrRNA-seq studies.