Increasing the number of classified taxa by 21.3% using SRA samples

Recently, we announced an increase in the classification rate of a human metagenomic sample from 74% to 97% by adding randomly selected SRA samples to the reference library. However, that result was flawed, most likely because of misannotated data which led to a high number of reads classified as “cellular organisms”. By using new SRA samples from the Logan project, we have now been able to avoid this problem and, for the same human microbiome subject sample, increase the number of classified taxa (at species/strain level) by 21.3%.

First we should describe what Logan is. Logan is a dataset recently released by a group of researchers (BioRXiv). Every sample in the Sequencing Read Archive (SRA) has been k-mer filtered (all 31-mers present with a count of less than 2 removed) and the result has been assembled into unitigs and contigs. This dataset represents essentially all publicly available NGS data held by the NCBI. By performing these operations of k-mer filtering and assembly, the size of the dataset has been reduced by about 40x, as a lot of redundancy and noise has been removed. This is an excellent starting point for further analyis.

We have carried out a series of experiments combining Logan samples with the standard Kraken 2 library (as well as with all of RefSeq) using Slacken. By adding 10,000 samples (bacteria/archaea/viruses) we increased the number of detected taxa from 665 to 719 in the randomly selected sample SRR31707655. The classification rate went up from 74.52% to 76.24%.

By adding 100,000 samples, the number of detected taxa went up to 807 (an increase of 21.3%) and the classification rate went up to 87.76%. The full results are available here (standard library) and here (Standard with 100,000 SRA samples).

The next step is to build a library with all bacterial/archaea/virus samples (approximately 3,000,000) as well as explore the effect of adding larger genomes, like eukaryotic samples. We have seen that adding more samples increases both the classification rate and the number of taxa detected. These libraries are also very cheap to build and operate, thanks to Slacken’s efficient algorithms. For the right domains, we believe that this technology could unlock substantial new discoveries and analytic value.

We are actively looking for commercial or academic partners and collaborators to help explore the next stage of Slacken work and push the possibilities of metagenomic profiling. If you think this might be you, please get in touch with us at: info@jnpsolutions.io