novoSplice: A gene, splice and quality aware RNA-seq aligner

novoSplice is an RNA-seq aligner which utilizes genes sequences as searching windows. The reduced indexing approach allows for an efficient, on the fly, hash-based indexing of genes, and it enables novoSplice to perform complicated splice junction identification, count features while performing reads alignment, and differentially penalize alignments depending on the gene bio-type. Reads bases qualities are utilized in the read alignment process and the refinement of the alignment score. To reduce reference bias, we implemented support for ambiguous (IUPAC) bases during indexing. In mapping single/paired-end RNA-seq reads, novoSplice implements a two-pass approach where a seed-and-vote algorithm is invoked to find a set of candidate genes. Then for each hit, a fast splice-aware chaining algorithm is called to obtain the best reads sequence alignment. In the event where the chaining process failed, a modified splice-aware Needleman–Wunsch algorithm is utilized to produce the final alignment. We benchmarked novoSplice against other state-of-the-art RNA-seq aligners according to SimBA [1] . SimBA benchmarking focuses on reads mapping to genome, splice junction alignment, single nucleotide variants and indels, and fusion based on simulated human RNA-seq (101bp paired-end) for normal and somatic mutation. We extended the scope to include additional RNA-seq aligners and simulated RNA-seq reads (100bp and 150bp paired-end) from multiple species (Homo sapiens, Mus musculus, Danio rerio, Drosophila melanogaster, Caenorhabditis elegans, Arabidopsis thaliana, and Saccharomyces cerevisiae). Currently, we are performing benchmarking on five aligners; novoSplice, novoAlign, STAR [2] , HISAT2 [3] and GSNAP [4] . In our preliminary benchmarking, and using SimBA somatic and normal 101bp datasets, novoSplice v0.4.1 shows consistent leading performance in both somatic and normal datasets. Extending SimBA study to include more data-sets from different species and read lengths would allow us to further investigate novoSplice performance in comparison to other aligners within the limits of simulated reads. novoSplice is in its beta phase and it is currently undergoing further performance tuning exercise for paired-end/single-end short reads routines. Support for long reads and single-cell RNA-seq and additional features; eg. detection of gene fusions, back-splicing, and trans-splicing will be available in the near future. novoSplice will be available at http://www.novocraft.com/products/novosplice for both Linux and macOS platforms.