Sunday, January 19, 2020

10X Genomics Exits Genomes

10X Genomics original product was a kit for generating linked reads from genomic DNA.  The idea had been kicking around for a while, partitioning long DNA into compartments and generating tagged libraries from each compartment.  This enabled both genome assembly and haplotyping from very small amounts of DNA.  When first reviewing 10X's slides from J.P. Morgan I had this thought "where's the genome kits" but then forgot to include it in my write-up.  Now I'm even more chagrined to discover that the explanation had been posted days before the conference: 10X has told their customers that their genome library kits are in the process of being discontinued.
The news broke on Twitter; I missed in the first time but then it popped in my feed some days later


The discontinuation is presumably a side effect of 10X's recurring defeats in patent battles with BioRad over the technology used in their early products.  When deciding what to reformulate into their newer formats that 10X believes are clear of BioRad's intellectual property, they have apparently decided to not go forward with the genome sequencing kits.

I had sometimes kept tabs on de novo assemblies of large genomes made with 10X libraries. While it was challenging to find ideal pairs of projects for comparison, the general impression I had is that 10X could do a good job, but the scaffold N50 for a 10X-boosted genome assembly would typically be in the range of the contig N50 for a true long read (PacBio or Oxford Nanopore) genome.  So all other things being the same, you'd have higher overall contiguity using true long reads.

Of course, cost, workflow and input material can be very important considerations.  With Sequel II on the PacBio side and PromethION for Oxford, the cost advantage of linked reads on Illumina has been shrinking.  For workflow, the overall workflow on the library prep is probably not radically different, though not being in the lab I'm not the best source on this.  10X did require only 10 nanograms of input, but now PacBio has an ultra low input prep down in that size range.  Both of those companies have de novo genomes and haplotyping as primary foci; these platforms are boosting performance at a rapid clip.

Perhaps also influencing 10X's decision is the launch of TELL-Seq from Universal Sequencing Technology.  This is a kit that claims to generate linked reads from DNA samples using bead-bound transposases, but without any droplets or specialized hardware.  If this really works nearly as well as 10X on functional criteria, then TELL-Seq would be very difficult to compete with on cost and workflow: it's a very simple Nextera-style workflow potentially amenable to automation and with no capital costs.

There are companies with similar products in the wings for single cell transcriptomics, so perhaps 10X decided to sacrifice a niche product with limited growth potential in favor of focusing on their core.  They also want to be the dominant player in spatial transcriptomics, and there are lots of potential competitors there.  So an apparent choice to focus at the expense of their oldest product line.

3 comments:

Steve Huang said...

There has been a "war" (a nice one, constructive) going on with long reads. Canu, Peregrine, Shasta, Flye, wtdbg2, just to name the most known ones (all came out in the past two years, or underwent major updates). NG50 based on the long reads all squash that of linked reads assemblies. Of course NG50 is not the only metric that matters.

And just look at that HiCanu assembly results shared last week at PAG. Just Wow!

Anonymous said...

There has been a "war" (a nice one, constructive) going on with long reads. Canu, Peregrine, Shasta, Flye, wtdbg2, just to name the most known ones (all came out in the past two years, or underwent major updates). NG50 based on the long reads all squash that of linked reads assemblies. Of course NG50 is not the only metric that matters.

And just look at that HiCanu assembly results shared last week at PAG. Just Wow!

Lutz said...

When starting with very high-quality DNA samples, 10X Genomics Genome library preparations result, in our hands, in Supernova de novo genome assemblies with scaffolds comprising entire chromosome arms -- for mammalian, bird, and reptile samples. No additional data type required. Thus, we could generate a high-quality mammalian genome assembly for $3500.
Scaffolding accuracy was certainly not perfect, but in relationship to the cost, the assemblies are amazing. Fish and plant samples are way more hit and miss affairs.