Mike Vella
Head of Machine Learning at Oxford Nanopore, driving fast algorithms and research to analyze DNA/RNA/Proteins.
- Yesterday in Malta the season for Dorado (Lampuki in Maltese) fishing opened, which is fantastic luck for me as I can enjoy a delicious dinner! My favourite basecaller and favourite fish share the same name.
- In this blog post, we show how chromatin accessibility can be profiled with minimal processing using nanopore sequencing – preserving native structure and capturing rich signal. Read it here: epi2me.nanoporetech.com/chromatin-ac...
- PolyA tail lengths can be accurately estimated using Nanopore dRNA
- We have developed workflows for Plasmid precursor quality control
- Oxford Nanopore simplifies RNA QC in GMP
- GMP QC is laborious and complex
- mRNA shows huge promise in personalised cancer vaccines
- Next up - Chris Alder discusses QC workflows in biopharma
- New for 2025 in EPI2ME - Teloseq, trio and pharmacogenomics workflows
- Single cell workflows now adds the ability to detect gene fusions
- Taxonomic classification
- EPI2ME workflows are next flow based you can upload your own
- 2ME format allows users to share workflows offline - code available on GitHub
- EPI2ME free to download and runs on Windows, Mac and Linux
- Sarah Grifftihs next up with updates on EIPI2ME
- We observe SOTA results in INDEL F1 scores for GIAB benchmarks, particularly big improvements in long INDELs
- Dorado variant is a completely new architecture - not a port of an existing variant caller
- Dorado v1.0.0 includes a preview release of our very own variant caller
- Dorado polish goes hand in hand with hifiasm- >2Q improvement to assemblies
- Dorado polish uses much more information than before, leading to higher quality assembly polishing.
- LSK-only Nanopore reads able to produce 23 gapless chromosomes , 10 scaffolded, and N50>130Mb
- Huge strides have been made in assembly. Nanopore-only T2T assembly is now feasible and will soon be routine.
- Many ways of answering the question “what is the genotype of my sample?”
- Up next - @rwinds.bsky.social discusses the latest in variant calling and assembly
- Base calling obeys scaling laws in deep learning - a Q28 base caller using standard chemistry is now available on demand.
- Base calling accuracy is compute-limited. As GPUs get faster so does accuracy!
- Dorado v1.0.0 introduces dramatic improvement to modified base calling speed
- Big improvements in modification calling accuracy - 6mA increases from Q16 to Q24
- New HAC models are 1.4Q more accurate - a 25% error reduction
- Dorado is no longer “just” a base caller - now includes alignment, mod base calling, alignment, and now even variant calling
- First up, @iiseymour.bsky.social with updates on all the new features in Dorado
- Up next at #nanoporeconf - Data for Lunch
- Reposted by Mike VellaI re-ran previous ONT basecalling benchmarks using the latest version of Dorado (thanks, @vellamike.bsky.social), and the RTX 40 series GPUs show an impressive speed boost! Check out the updated benchmark results at the bottom of this post: rrwick.github.io/2024/08/16/s...
- Dorado now supports bacterial genome polishing. It can output assemblies in FASTQ, including Q scores. @rrwick.bsky.social 's analysis shows Q scores are well correlated with base call reliability, meaning they can help reduce false positives in variant calling. rrwick.github.io/2025/02/19/f...
- Reposted by Mike VellaAwesome to see the assembly polishing models we've been working on the last couple of months finally out there - we've had some really promising results with these internally so keen to see how others get on with them!
- 🎄 New Dorado release (v0.9.0) – and it's big! 🚀 Dorado polish command (experimental) for improving draft assemblies – faster & more accurate than Medaka ⚡ Faster modified base calling models 🔧 Usability & accuracy improvements: PolyA, Barcoding, 6mA calling github.com/nanoporetech...
- 🎄 New Dorado release (v0.9.0) – and it's big! 🚀 Dorado polish command (experimental) for improving draft assemblies – faster & more accurate than Medaka ⚡ Faster modified base calling models 🔧 Usability & accuracy improvements: PolyA, Barcoding, 6mA calling github.com/nanoporetech...
- Reposted by Mike VellaThis is worth looking at. Trying some genomes we have assembled with earlier versions of hifiasm, hifiasm plus herro and then this new version of hifiasm. The early assembly results suggest that now hifiasm alone is equivalent to what we got with herro but with less compute!
- Exciting news! The latest hifiasm release from Haoyu Cheng and Heng Li adds beta support for @nanoporetech.com simplex R10 reads. Initial results look very promising. 🚀 Check it out: github.com/chhylp123/hi...
- Exciting news! The latest hifiasm release from Haoyu Cheng and Heng Li adds beta support for @nanoporetech.com simplex R10 reads. Initial results look very promising. 🚀 Check it out: github.com/chhylp123/hi...
- Reposted by Mike VellaExcited to join the #Nanopore Community on BlueSky. Follow us for the latest updates and discussions about Oxford Nanopore and Nanopore Community. #WYMM
- At @nanoporetech.com, we've used ML to achieve remarkable advancements in basecalling. Now, we're focusing on driving further improvements in consensus and variant calling, enhancing both speed and accuracy. Watch @rwinds.bsky.social update on our progress: www.youtube.com/watch?v=IB6D...
- Just released an update to ONT's basecaller Dorado (version [v0.4.3]) : 🧬 🖥️ * m6A DRACH-context mod model * poly(A) / poly(T) tail estimation * RNA read splitting * RNA adapter trimming Plus sample sheet support * Significant memory reduction for duplex calling. github.com/nanoporetech...
- This paper is amazing. Using Oxford Nanopore sequencing in CNS tumor diagnostics is a remarkable example of engineering meeting clinical innovation. Our work on accurate GPU-accelerated basecalling has helped enable sample-to-sequence, with methylation, swiftly. www.nature.com/articles/s41...
- Reposted by Mike VellaCNS tumor classification during surgery using rapid Oxford Nanopore sequencing. This technology is so promising for real-time, near-patient genomic analysis, not least because of direct DNA methylation detection. Very exciting paper - congrats! www.nature.com/articles/s41...
- Release of v1.0 of the HG002 genome benchmark represents an important milestone. The Q100 project's aim to establish a complete and perfectly accurate diploid HG002 genome as a new standard is commendable. These developments are advancing the precision of tools and sequencing technologies.
- We are excited to release v1.0 of our HG002 genome benchmark today! Our goal with the Q100 project is to create a complete and perfectly accurate representation of the diploid HG002 genome to serve as a replacement for current variant-based benchmarks. More info here: github.com/marbl/HG002
