Multi-modal benchmarking of the Ultima UG100 and Illumina NovaSeq sequencing platforms using clinically relevant FFPE tissues

This study demonstrates that the Ultima UG100 sequencing platform delivers performance comparable to Illumina NovaSeq across multiple genomic assays using FFPE tissues, offering a high-precision, cost-effective alternative for clinical and population-scale research despite minor, non-critical technical differences.

The Gist

Imagine you are a detective trying to solve a complex crime. The "crime scene" is a patient's tissue sample (specifically, old, preserved tissue blocks known as FFPE, which are like crime scene photos taken years ago). To solve the case, you need to read the "diary" written inside the cells of that tissue. This diary is made of DNA and RNA, and reading it requires a high-tech machine called a sequencer.

For a long time, there has been only one reliable detective agency in town: Illumina. Their machines are the gold standard. They are expensive, but everyone trusts them because they rarely make mistakes.

Recently, a new, budget-friendly detective agency opened up called Ultima Genomics. They use a different, faster, and cheaper method to read the diary. They claim they can do the job for a fraction of the cost. But the big question remains: Can this new, cheaper agency read the diary just as accurately as the old, expensive one, especially when the diary pages are old and slightly damaged (like FFPE tissue)?

This paper is the result of a massive "taste test" or "head-to-head competition" between these two agencies.

The Setup: A Double-Blind Test

The researchers took 15 different tissue samples from patients with various diseases (like different types of cancer and inflammatory bowel disease). They split the samples in half.

• Team A sent one half to the Illumina lab.
• Team B sent the other half to the Ultima lab.

They tested four different ways of reading the diary:

1. Whole Genome (WGS): Reading every single letter of the DNA.
2. Whole Exome (WES): Reading only the important chapters (the parts that make proteins).
3. Whole Transcriptome (WTS): Reading the "active notes" the cells are currently writing (RNA).
4. Single-Nucleus (snRNA-seq): Reading the notes from individual cells, one by one, to see who is in the room.

The Results: How Did They Compare?

1. The "Typo" Problem (Errors)

Every typewriter makes mistakes.

• Illumina is like a very careful, slow typist. They almost never miss a letter, but they sometimes get confused when two words look very similar (like "form" and "from"). They tend to be very sensitive, catching even the faintest whispers, but sometimes they hear things that aren't there (false alarms).
• Ultima is like a fast typist who uses a different ink. They are generally very accurate, but they have a specific quirk: they sometimes accidentally add an extra letter when typing a long string of the same letter (like typing "baaaad" instead of "bad"). This is called an "indel." However, the researchers found that Ultima is actually better at ignoring the "ghost whispers" (false alarms) that Illumina sometimes catches. Ultima is more conservative; it only reports what it is 100% sure of.

2. The "Lost Pages" (Missing Data)

In the RNA tests (reading the active notes), the researchers found that Illumina was slightly better at finding very rare, obscure notes (like pseudogenes, which are broken copies of genes). Ultima missed a few of these obscure notes.

• The Analogy: Imagine reading a library. Illumina finds every single book, even the dusty, forgotten ones in the basement. Ultima finds all the popular, important books and the main storylines perfectly, but might miss a few of the obscure, dusty pamphlets.
• The Verdict: For the main story of the disease (the cancer drivers, the immune response), both machines told the exact same story. The missing pamphlets didn't change the plot.

3. The "Cell Count" (Single-Cell Analysis)

When they looked at individual cells, both machines agreed on who was in the room. They identified the same types of immune cells and cancer cells.

• The Analogy: If you take a photo of a crowd, both cameras captured the same people in the same groups. The only difference was that the Ultima camera took a slightly higher-resolution photo (more pixels), but it didn't actually see more people; it just took more pictures of the same people.

4. The "Cost vs. Quality" Trade-off

The study confirmed that Ultima is indeed much cheaper.

• Illumina is like a luxury car: smooth, reliable, and everyone knows how to drive it. It catches every little detail, even if some are just noise.
• Ultima is like a new, high-tech electric car: it's faster, cheaper to run, and gets you to the same destination. It has a slightly different engine sound (different error patterns), but it gets you there just as safely for the main journey.

The Big Conclusion

The researchers concluded that Ultima is ready for prime time.

While Illumina is still the "gold standard" for regulatory approval (like getting a license to drive), Ultima is now proven to be a viable alternative for large-scale research. If you want to study thousands of patients to train an AI to predict diseases, you don't need to spend a fortune on Illumina. You can use Ultima, save a massive amount of money, and still get the same biological answers.

The Takeaway Metaphor:
Imagine you are trying to map a new city.

• Illumina sends out a team of surveyors who walk every single street, measuring every crack in the sidewalk. It's perfect, but it takes a long time and costs a fortune.
• Ultima sends out a drone fleet. They fly over the city faster and cheaper. They might miss a tiny crack in a specific sidewalk (a rare gene), but they map the highways, the neighborhoods, and the traffic patterns perfectly.

For the purpose of understanding the city's layout (the disease), the drone map is just as good as the walking survey. This study proves that we can now use the "drones" (Ultima) to map the human genome on a massive scale, making medical research cheaper and faster for everyone.

Technical Summary

1. Problem Statement

High-throughput sequencing is essential for precision medicine, drug discovery, and AI-driven multi-omic analysis. While Illumina's Sequencing-by-Synthesis (SBS) technology has long been the industry standard, emerging platforms like Ultima Genomics' UG100 (utilizing mostly natural sequencing-by-synthesis, or mnSBS) promise significantly lower costs and higher scalability. However, the performance of these new chemistries on archival clinical samples (FFPE)—which are often degraded and critical for retrospective studies—remains poorly characterized. There is a lack of comprehensive, multi-modal benchmarking to determine if Ultima's data is biologically equivalent to Illumina's across different genomic modalities (RNA-seq, WES, WGS) and whether it can be integrated into existing clinical and research workflows without compromising biological validity.

2. Methodology

The study conducted a rigorous, head-to-head comparison of the Ultima UG100 and Illumina NovaSeq platforms using a diverse cohort of 15 FFPE samples from five disease indications:

• Oncology: Diffuse Large B-cell Lymphoma (DLBCL), Glioblastoma (GBM), Muscle-Invasive Bladder Cancer (MIBC).
• Immune-mediated: Ulcerative Colitis (UC) and Crohn's Disease (CD).

Experimental Design:

• Modalities: Single-nucleus RNA-seq (snRNA-seq), Whole-Transcriptome Sequencing (WTS), Whole-Exome Sequencing (WES), and Whole-Genome Sequencing (WGS).
• Workflow:
  • Illumina: Libraries prepared and sequenced at CeGaT (Germany) using NovaSeq 6000/6000X Plus. Data processed via the DRAGEN pipeline.
  • Ultima: Identical library aliquots were converted with Ultima-specific adapters and sequenced on the UG100 (Fremont, USA). Data processed via DeepVariant (optimized for Ultima).
• Normalization: To ensure fair comparison, Ultima datasets were downsampled to match Illumina read counts.
• Analysis: Metrics included sequencing quality, coverage uniformity, error spectra, variant concordance, gene expression reproducibility, and cell-type clustering.

3. Key Contributions

• First Comprehensive FFPE Benchmark: This is the first study to evaluate the UG100 platform across four distinct omics modalities using clinically relevant, archival FFPE tissues.
• Multi-Modal Validation: It moves beyond simple WGS benchmarking to assess transcriptomics (bulk and single-nucleus) and exomics in complex disease contexts.
• Pipeline vs. Chemistry Disentanglement: The study highlights that many observed differences are driven by the combination of sequencing chemistry and the specific bioinformatic pipelines (DRAGEN vs. DeepVariant) rather than the chemistry alone.
• Clinical Utility Assessment: It specifically evaluates the detection of clinically actionable variants (oncogenic hotspots, tumor suppressors) and disease-associated germline variants (IBD GWAS loci).

4. Key Results

A. Bulk RNA-seq (WTS)

• Expression Concordance: Both platforms showed high correlation in gene expression ($r \approx 0.95-0.96$) and captured key oncogenic and immune-related transcripts with similar sensitivity.
• Biotype Differences: Illumina detected slightly more pseudogenes and non-coding RNAs, likely due to shorter read lengths mapping to homologous regions. Ultima showed a slight enrichment for protein-coding genes, possibly due to longer reads providing better spatial context for unique mapping.
• Biological Impact: Despite minor differences in individual gene detection (e.g., some oncogenes like TDGF1 were detected by Illumina but missed by Ultima), pathway-level conclusions remained identical. No batch correction was required to integrate data for biological signal preservation.

B. Single-Nucleus RNA-seq (snRNA-seq)

• Cell Clustering: Both platforms recovered major cell types with nearly identical clustering patterns. Cells clustered by biological cell type rather than sequencing technology.
• Depth vs. Quality: Ultima generated ~5x more reads per cell, but this resulted in oversampling rather than increased transcript diversity (UMI counts were similar).
• Exceptions: Minor deviations were observed in the DLBCL cohort regarding B-cell vs. T-cell proportions, likely due to differences in UMI capture efficiency or thresholding.

C. Whole-Exome Sequencing (WES)

• Coverage: Both achieved high target coverage (>95% at $\ge 20\times$).
• Variant Calling Strategy:
  • Ultima (DeepVariant): Produced a conservative callset with high precision but fewer unique calls. It prioritized high-confidence variants, resulting in fewer low-frequency artifacts.
  • Illumina (DRAGEN): Produced a permissive callset with higher sensitivity, detecting many low-VAF (<10%) variants that were often low-confidence or artifacts.
• Clinical Actionability: Both platforms reliably detected known oncogenic hotspots (e.g., TP53, PTEN, EZH2) and tumor suppressor mutations.
• Error Profiles: Ultima showed a characteristic bias toward indels in low-complexity regions (homopolymers), whereas Illumina had uniformly low indel rates.

D. Whole-Genome Sequencing (WGS)

• Germline Variants: Ultima demonstrated higher sensitivity for true germline variants but a significantly higher false-positive burden for indels compared to Illumina when benchmarked against the GIAB truth set.
• Artifact Reduction: Crucially, Ultima generated far fewer spurious low-frequency artifacts (low-VAF variants) compared to Illumina. In the IBD cohort, Illumina's low-frequency calls were ~70% of its total callset, whereas Ultima's were only ~6%.
• GWAS Variants: Illumina detected a larger fraction of curated IBD-associated GWAS variants, likely due to its permissive calling thresholds, though all variants detected by Ultima were also found by Illumina.

5. Significance and Conclusion

• Translational Viability: The study confirms that the Ultima UG100 platform is a viable alternative to Illumina for clinical and translational research using FFPE samples. It captures biologically meaningful signals (cell types, pathways, driver mutations) with high fidelity.
• Cost-Efficiency: The platform offers a pathway to reduce the cost of large-scale genomic studies (potentially <$0.01/cell for snRNA-seq) without sacrificing biological insight.
• Data Integration: Data from both platforms can be combined for meta-analysis, provided that platform-specific analytical considerations (e.g., variant calling stringency, indel error modes in homopolymers) are transparently addressed and calibrated.
• Future Outlook: While the UG100 shows a "cleaner" but more conservative profile, the authors note that the newer UG200 series includes workflow improvements that may further narrow performance gaps. The study emphasizes that for AI-driven multi-omic modeling, the stability and reproducibility of biological signals are paramount, and Ultima meets these criteria despite differences in raw base-call error profiles.

Limitations: The study used a modest cohort size and relied on platform-optimized pipelines (conflating chemistry and workflow effects). Future work requires larger cohorts, fresh-frozen tissue comparisons, and orthogonal validation to fully characterize rare variant sensitivity.