About

The assembly pipeline

Semblans (pronounced “semblance”) is a C++ tool that abstracts the quality-control, reconstitution, and post-processing steps of transcriptome assembly from raw short-read sequences to annotated coding sequences. It leverages C++ data streaming to pass large sequence data between successive program calls without costly reformatting on disk, and it has been tested across a range of computing architectures including the NSF ACCESS cyberinfrastructure ecosystem.

The pipeline is organised into three successive phases, each of which can be run individually or in tandem:

1. Preprocessing. Reads are first screened with FastQC. Single-base sequencing errors are corrected with Rcorrector (reads flagged as “unfixable” are removed), synthetic adapter sequences are trimmed with Trimmomatic when detected, foreign reads are filtered out via the k-mer-based taxonomic classification of Kraken2, and overrepresented sequences identified by a second FastQC pass are removed.
2. Assembly. Cleaned reads are assembled de novo into a preliminary set of contigs with Trinity.
3. Postprocessing. Chimeric contigs are detected via a BLASTX-compatible DIAMOND alignment against a user-supplied reference proteome and removed following the Yang & Smith (2013) procedure. Read support is quantified with Salmon, transcripts are clustered with Corset to extract a gene-level approximation of the data, and a final BLASTP-compatible DIAMOND alignment guides TransDecoder in predicting the final coding regions and peptides. Optional functional annotation is performed with HMMER against the PANTHER peptide database.

In benchmarking, Semblans produced higher-quality assemblies — measured by TransRate score — for 98 of the 101 short-read runs tested, drawn from the 1KP project and from a Caryophyllales-focused study.

For documentation, source code, and binaries see the project on GitHub and the project wiki.