Isopod: detecting differential isoform usage from long-read single-cell data

Abstract

A major limitation of short-read, single-cell sequencing is the inability to reconstruct full-length transcripts, hindering isoform-level investigation at the single-cell level. Nanopore sequencing can be used to sequence transcripts captured by the 10x single-cell platform, allowing us to study not only gene expression differences between cell types but also isoform-specific differences. In addition, isoform-level data contains the information to detect isoform switching between cell types, but currently there are few established methods for this analysis. We aim to utilise the strengths of long-read, single-cell data by presenting a permutation-based method to estimate differential isoform usage between cell types or cell clusters. Our method, Isopod, takes an isoform-level counts table from a sample of single cells. Isoforms are first filtered, discarding low-count and sparsely represented genes, and then low-count isoforms within genes are collapsed, reducing unwanted isoform variation caused by misassignment of noisy reads. The cell-by-isoform counts table is then reduced to a contingency table of pseudo-bulk counts that tests for changes in the proportion of an isoform within a gene in a designated cell type compared with all other cell types. The permutation function then randomly shuffles cells between the cell clusters and the test is regenerated. This is repeated for 10,000 permutations to calculate an empirical permutation p-value, identifying changes in isoforms proportions between cell-type clusters. Comparing Isopod to a chi-squared analysis without permutation revealed significantly fewer false positives.

Biography

Michael Nakai completed his undergraduate degree in pathology at the University of Melbourne, with an honour’s year focused on creating nanoparticles that targeted cancer cells. He then worked as a Research Assistant, investigating the interplay between the gut microbiome and blood pressure at Monash University, where he started in the wet lab. To analyse the genomic data collected from the gut microbiome, Michael began transitioning into bioinformatics. Following three years as a Research Assistant, Michael started a PhD with Alicia Oshlack at the Peter MacCallum Cancer Centre, and now works on long-read, single-cell transcriptomics data.