February 27, 2024 | Among the sequencing news at AGBT earlier this month, Ultima Genomics announced the commercial launch of its UG 100 ultra-high throughput sequencing platform.
The UG 100 system has been in an early access program and is now the first commercial product for Ultima, a company founded in 2016 and currently with $600M in funding from Khosla Ventures, Andreessen Horowitz, and more. Ultima’s CEO, Gilad Almogy, has a PhD in applied physics from California Institute of Technology. Before Ultima, he founded a solar energy company, and before that he worked with semiconductors. Ultima’s approach to sequencing draws directly from Almogy’s history. Ultima is conducting sequencing on silicon wafers and adding reagents via spin coating.
“Fundamentally, everyone for the last ‘forever’ has been using flow cells, and the flow cell has a finite physical size,” Almogy pointed out in a conversation with Bio-IT World. “Scaling the flow cell turns out to be very expensive, both because the materials of which it’s made, and the process you put into it… That kind of constrains how much data you can get.” Using an open silicon wafer, however, is cheap and scalable.
The Ultima platform “kind of piggybacked on this very large, very accurate surface, which is the silicon wafer,” Almogy said, in the 200mm diameter wafer size. “They cost nothing, or very, very little,” he added. “This wafer is patterned at micron scale generating a dense array of electrostatic landing pads to bind clonally amplified sequencing beads, which are separately produced by an automated emulsion PCR process,” the company shares in a technology whitepaper available on BioRXiv. To add reagents, Ultima uses spin coating, dripping fluid into the center of a spinning wafer to quickly achieve a uniform film coating the wafer.
The use of the silicon wafer and spin coating reagents are the two advantages have driven the Ultima Genomics cost per genome to about $100, Almogy explained. “We achieved the $100 genome without saying, ‘Oh, you can achieve $100 genome if you pull 1,000 genomes in a $100,000 run.’ No. It’s actually a $2,000 run.”
The sequencing itself is a massively parallel sequencing-by-synthesis (SBS) approach. Mostly natural sequencing-by-synthesis (mnSBS) uses in each sequencing cycle a single base from a mostly natural nucleotide (MNN) mix, which comprises a minority (<20%) of fluorescently labeled, non-terminated nucleotides and a majority of unlabeled, non-terminated (natural) nucleotides, the authors write in the technology paper. The resulting synthesized DNA is mostly unmodified, yet incorporated bases can still be efficiently measured at the reaction endpoint via high throughput optical scanning with a high signal-to-noise ratio, based on hundreds of photons produced by each fluorescent label.
“It’s essentially an optical system,” Almogy explained. High dynamic range cameras photograph the silicon wafers between reagent steps, then all fluorescent labels are cleaved and washed, and the wafer cycles back to reagents. Scanning of the entire surface of the wafer generates non-overlapping tile images which are processed using on-board NVIDIA GPUs to extract bead locations and corresponding raw signal vector per clonally amplified sequencing bead over multiple sequencing cycles. A deep convolutional neural network (CNN) converts raw signals into sequence reads. “We built it up on a machine learning stack from day one,” Almogy said, “a kind of GPU and NVIDIA-based machine learning stack which could deal with all of the [complexity].”
Ultima is not interested in “democratizing sequencing” in the historic sense, meaning to put a small, expensive sequencer at every researcher’s desk. “It’ll be a large box, fully automated, that will give you very, very low-cost data,” Almogy explained, adding that the UG 100 is a bit larger than a NovaSeq.
Each run lasts 12 to 14 hours for two wafers cycling between the chemistry and imaging steps. Six wafers can be loaded on the instrument at a time; reagents are stored in bulk tanks to dispense as needed. “And the tool just runs. So if the tool finishes a run on Sunday at 3am, it just kicks off the next run. It’s an automated tool,” Almogy explained.
The system is designed for large academic labs, CLIA-certified labs, large genome centers, and oncology and liquid biopsy companies, Almogy said. “There’s folks offering therapy selection for tumors… You see a wave taking off in the last year or two of MRD—minimal residual disease—[testing], and of course screening, where it’s specific tumor screening or multi-cancer early infection.”
All of these use cases are very different from germline sequencing, he said, because of the concentration of DNA in the sample. “Now you need two things: you need to be able to read a lot of DNA cost effectively, and you need to trust a single read.”
Ultima is reporting high base accuracy (Q30 > 85%), high accuracy for SNPs (99.6%), SNV F1 of 99.8%, and INDEL F1 of 99.4%. “The main accuracy gaps that remain are strongly correlated with repetitive regions and extreme %GC regions that are difficult to amplify efficiently and can likely be minimized by additional optimization of the clonal amplification protocol,” the authors write in their technical paper. “Indels in homopolymer regions, which are traditionally the Achilles heel of non-terminating chemistries, can be called with good accuracy up to lengths of 8-10 bases. Further improvements in features such as polymerase fidelity, base calling, and variant calling algorithms, are expected to continue driving overall accuracy gains over time.”
Ultima also includes a very high accuracy feature called ppmSeq, paired plus minus sequencing, that is focused on rare event detection. Keith Robison, author of the Omics Omics Blog, got to see the Ultima in a preview event, and he noted that ppmSeq seems to be the feature generating the most excitement right now. “The advantage of Q60 reads for needle-in-a-haystack problems such as detecting oncomutations in cell-free DNA is easy to grasp,” he writes in his AGBT wrap up blog.
Almogy reported 12 Early Access users that he calls the “high quality, biggest users of sequencing in the world.” Those publicly mentioned mentioned include the Broad Institute, the New York Genome Center, Regeneron, Hebrew University, Baylor College of Medicine, Exact Biosciences, and Quest Diagnostics, with whom the company has announced a collaboration. As far as longer term buzz, now we watch and wait.