RNA-Bloom provides lightweight reference-free transcriptome assembly for single cells

Abstract

We present RNA-Bloom, a de novo RNA-seq assembly algorithm that leverages the rich information content in single-cell transcriptome sequencing (scRNA-seq) data to reconstruct cell-specific isoforms. We benchmark RNA-Bloom's performance against leading bulk RNA-seq assembly approaches, and illustrate its utility in detecting cellspecific gene fusion events using sequencing data from HiSeq-4000 and BGISEQ-500 platforms. We expect RNA-Bloom to boost the utility of scRNA-seq data, expanding what is informatically accessible now.The early stages of single-cell RNA sequencing (scRNA-seq) data analyses have been primarily limited to gene expression quantification and individual splice-junction detection 1 .Isoform-level analyses for single cells remain scarce 2 . Although long-read sequencing technologies excel in capturing near full-length transcript sequences, which are ideal for isoformlevel analyses 2 , they have a significantly higher error-rate, limited sequencing depth, and higher input requirement compared to short-read sequencing technologies 3 . Short-read sequencing prepared with transcript-end capture protocols 4-6 are scalable to nearly one million cells, but they exhibit strong transcript-end bias, prohibiting the reconstruction of alternative isoforms. Lifting this bias, there are other short read sequencing protocols 7, 8 that offer full-length transcript

Terpene Synthases and Terpene Variation in Cannabis sativa

“…While separate transcriptomes do not permit quantitative comparison between cultivars, they eliminate the risk of quantitation errors from mapping kmers from similar transcripts between cultivars. We used the RNA-Bloom assembler (Nip et al, 2019), designed for single-cell RNA-seq libraries, to capture the diversity of sequences across the five cultivars while reducing the possibility of chimeric contigs. Contigs with .98% predicted amino acid sequence identity were collapsed under the longest representative sequence.…”

“…All merged and unmerged reads were pooled and first assembled with Trinity (version 2.6.5) to generate 599,285 nonredundant contigs with an average length of 511 bp. To gain insight into cultivar-specific sequences, we reassembled all of the unmerged sequences using RNA-Bloom (version 0.9.8;Nip et al, 2019) to generate five separate assemblies, one per cultivar, with an average of 260,000 nonredundant contigs and average length of 1,400 bp.…”

Complete Biosynthesis of the Anti-Diabetic Plant Metabolite Montbretin A

“…To identify the remaining UGTs required for the completion of MbA biosynthesis, we explored the RNA-Seq library prepared from yC (harvested June 10 th 2016) described in Irmisch et al (2018) and a newly generated set of five RNA-Seq libraries developed from yC harvested at five different time points (June 27 th , July 22 nd , August 16 th , September 12 th , and October 6 th 2016). Transcriptomes for each of these six time points of yC development were constructed using RNA-Bloom (Nip et al, 2019), predicted peptides were combined and redundancies reduced yielding 40,565 non-redundant (NR) transcript contigs covering translated sequences with an average length of 331 amino acids (aa). Using reciprocal BLASTP searches, 190 UGTs (≥ 250 aa) were identified in the montbretia yC-time course transcriptome (Supplemental Data Set 1).…”

Terminitor: Cleavage Site Prediction Using Deep Learning Models

“…Here we propose a Poly(A) site prediction pipeline based on Termin(A) n tor. The pipeline starts with RNA-Bloom, a fast and memory-efficient de novo transcriptome assembler (35). RNA-Bloom is run with the option -stratum 01 that allows extension of all fragments regardless of its coverage, and the option --polya that prioritizes the assembly of transcripts with poly(A) tails.…”