Setting Yourself Up For Success: Choosing The Right Omics Platform For Your Diagnostics Pipeline

Meeting where you are at. Growing with you.

By Darren Ames, Head of Solutions Science

The Aligner Is Never The Problem.

BWA-MEM hasn't personally betrayed anyone since roughly 2013. GATK is grumpy but honest. And yet at 2:14 a.m. on a Sunday, somebody's phone is going off, and it's never because a Smith-Waterman traceback lost its nerve. It's because the LIMS emitted a sample ID with a trailing whitespace character, the orchestrator read that as a new sample, and now there are two of them. One is a ghost, and the ghost has a requisition number attached to a real person waiting on a real result.

Slight tangent: I keep 26 chickens. They're all named—Verbena, Lars, LaFonda, Buttercup, Matilda, Bob, PB, Becky, Mini Becky, Corn, Big Chongus 2, and, because the naming committee (my 12-year-old daughter) operates without oversight, Uncorn. Becky and Mini Becky are entirely distinct individuals with vastly different opinions about mealworms. Hand my flock to an unconfigured LIMS and it would silently merge them, turning Mini Becky into a ghost.

The Rebuild Tax Nobody Budgets For

Diagnostics pipelines don't hold still. The lab validating its first assay this quarter might be running a nationwide menu in three years, or pivot twice getting there. That's the job. The trouble is that nearly every infrastructure decision forces a bad choice:

• Build for where you are: Scripts, a moody cron job, and an S3 bucket named by someone who left for a company with better snacks. It works beautifully until an auditor demands complete lineage for a run from 14 months ago, at which point "reproducible" and "we still have the Dockerfile somewhere, probably" turn out to be very different phrases.
• Overbuild for an unfunded future: Burn 18 months of precious runway building massive custom infrastructure that isn't your core product.

Both roads end at a re-platform. In a regulated environment, a re-platform isn't a migration—it's a revalidation. Every assay, every version, every piece of evidence, all over again. “One does not simply walk into Mordor” (re-platform a clinical lab). I've watched brilliant teams eat 6–9 months on this, and it never appears on a budget line item because nobody plans for it. They just collide with it.

Why Bet On DNAnexus

Look, I'm biased. I've worked here for more than a decade. But the math speaks for itself: 8 of the 10 largest diagnostics pipeline developers build on DNAnexus. We back 130+ enterprise customers, 60,000+ users across 48 countries, with 140+ petabytes under management. FedRAMP Moderate, turnkey GxP, ISO 27001, CAP/CLIA, HIPAA, GDPR—with institutions like City of Hope and Dana-Farber running real precision oncology, not sandbox pilots. Internal benchmarks consistently show ~50% lower cost per sample at scale and meaningfully faster validation cycles. Ask us to show our work; we will.

The proof point I'd bet money on, though, is the boring one. Point solutions force a re-platform at every growth inflection: the research tool that can't go clinical, the clinical stack that can't scale, or the production system with no room for R&D. Revalidations are where roadmaps go to die quietly. Most labs at the massive scale arrived there having started with us when they were considerably smaller.

Four Shapes Of The Same Clinical Diagnostics Platform

We've packaged the diagnostics side of our platform into a clear ladder—Starter, Emerging, Scaler, Enterprise—with a substantially lower entry point than before. I'm not going to talk pricing numbers here; that's a conversation with a human who is paid to have it. The interesting part is the shape.

All four tiers run on Titan, our core platform: same execution substrate, same governance engine, same audit spine, same compliance posture. These aren't four separate products. They're one function evaluated over a widening domain. Climbing from Starter to Enterprise isn't a migration—it's configuration and service hours. Your pipelines aren't rewritten, your audit history isn't orphaned, and your validation evidence doesn't become digital archaeology. You don't revalidate simply because you grew.

Two of those rows deserve extra focus:

• Audit trail on at Starter: Lineage cannot be generated retroactively. If run 4,112 happened 11 months ago and nobody captured which container digest, reference build, parameters, and inputs were used—it’s gone. You can't reconstruct lineage from vibes and a Slack thread. Tolkien wrote ninety pages of appendices establishing who begat whom because lineage is either recorded as it happens or invented later. Until someone audits you, your pipeline sits in superposition: reproducible and non-reproducible at once. Turning capture on while it costs virtually nothing is how you ensure the wave function collapses somewhere pleasant.
• Enterprise scale arithmetic: At scale, percentages change character. A 3% rerun rate at 500 samples a month is 15 reruns and a quick Slack message. At a million samples a year, 3% is 30,000 reruns—a headcount issue, a compute line item Finance flags, and a TAT tail long enough to breach commitments to real patients. Enterprise delivers reserved capacity because when you need 40,000 cores on the last Tuesday of the quarter, so does everyone else in your region.

Enterprise isn't a different platform. It's the same Titan a two-person startup gets on day one, just with reserved capacity and a dedicated engineering team wrapped around it. A startup at Starter isn't on a beginner tier it will outgrow and abandon; it's standing on an engine that has already survived industrial volume, with the heavy-duty valves turned off until needed.

The Unglamorous Conclusion

The goal of all of this is to make clinical informatics boring. Not simple—the science will never be simple. Boring in the sense pilots use the word: no surprises, everything right where you left it, the checklist works, and nobody's phone rings at 2:14 a.m. about a ghost sample.

One last confession, since I've mentioned rigorous unit economics: my personal chicken math doesn't add up. I spend materially more on feed, bedding, swings, toys, and heated waterers of questionable engineering than I will ever recover in eggs. But the eggs were never the point. Twenty-six small, loud, deeply goofy, loving, extremely alive things come running when I open the gate, and whatever that does for the human soul doesn't show up on a spreadsheet.

Measure what should be measured, ruthlessly—and know which things were never about the measurement. Cost per sample belongs in the first category. Why you got into this work belongs in the second.

So bring us your worst pipeline. Genuinely, the worst one. We will lift it for free. The gnarlier the source tree, the more interesting the report—and I have a feed bill to justify.