Transitioning from short to long read transcriptomics: accuracy, bias and analysis challenges

Recent advances in long-read sequencing technologies like PacBio and Oxford Nanopore have revolutionized the generation of full-length transcript sequences. These technologies facilitate a deeper understanding of complex isoforms and transcript structures. As the precision and depth of sequencing improve, long-read methods are becoming more prevalent in transcriptomics studies for identifying differential gene expression and isoform utilization across various conditions using multiple replicates. Concurrently, new algorithms for transcript reconstruction and quantification have emerged, adapting to the influx of long-read data. With the field's shift from short to long reads, there is an imperative to establish optimal data preprocessing, experimental designs, quantification, and normalization strategies tailored to these data types. Critical questions arise: What is the quality of my transcript identification and quantification calls using long-read transcriptomics data? What is the best approach for constructing a long-read-based quantification table? How many replicates are necessary? What is the ideal sequencing depth? How can one identify and correct potential biases in transcript quantification? Which data analysis strategies, if any, are different in long-read transcriptomics? Do these considerations vary depending on the chosen sequencing technology or the algorithm used for processing long reads? I will present the efforts from my lab to evaluate the quality and utilization of long-read transcriptomics data and discuss what challenges are still present to realize a complete shift from short to long reads in transcriptomics studies.