A Survival Guide to Oxford Nanopore Analysis

Tool documentation provided by the BatchX platform

We are living in unprecedented times. In less than 50 years, we have gone from developing the first DNA sequencing method and applying it in the Human Genome Project, which took 13 years to complete, to being able to complete the same in less than a day.

It all started with the Sanger DNA sequencing method, developed in 1977. Although fewer people realized it at the time, this achievement was as remarkable as the Moon landing 8 years prior. The door opened for molecular biologists to perform studies that had never been done before, such as the discovery of genes responsible for certain diseases (e.g., cystic fibrosis) or the complete sequencing of virus or bacterial genomes (e.g., H. influenzae).

Twenty years later, Next-Generation Sequencing (NGS) ushered in the so-called “golden age of genetics”, lowering sequencing costs and allowing labs, from small to large, to use these techniques. The impact of NGS technologies proved to be far reaching and enabled previously unimaginable scientific achievements and biological applications. For example, making human genome sequencing economically viable (the first draft of the human genome, developed during the Human Genome Project, cost $300 million) enabled screening the human population for the discovery of genetic variants, which in turn opened new fields of research (e.g., how does our genetic code predispose us to certain types of cancer). Illumina has been the flagship of this generation and is still the current market leader in sequencing. Over 80% of sequencing experiments are done with Illumina, representing yearly revenues of over $3 billion.

Third Generation Sequencing (TGS) has revolutionized the study of epigenetics, transcriptomics, metagenomics, and genomics by providing new and improved methods, to name a few. An achievement of this technology has been the sequencing of a whole chromosome (human chromosome X), without any gaps, for the first time. With the long reads generated with TGS techniques, large repetitive regions can be sequenced in one single read (see image below), therefore avoiding the problem of assembling short reads that have the same repeated information.

Advantage of long reads vs short reads in the assembly of repetitive regions. (Image Credits)

TGS produces much longer reads that alleviate the computational challenges that appear when building contigs that are much larger than the size of typical NGS reads. Typical Illumina (i.e., NGS) reads range from 50 bp to 300 bp, much shorter than the tens of thousands up to millions of base pairs that some TGS technologies can achieve, facilitating the sequencing of entire regions of repetitive sequence and structural variation in complex genomes [1]. Another advantage is that for epigenetic marker experiments NGS relies on additional library preparation steps while TGS can be performed directly. Furthermore, TGS allows for direct RNA sequencing as there is no need for conversion to cDNA, something that is required by traditional RNA sequencing techniques. [2, 3] Additionally, a lower upfront cost is required to begin sequencing with ONT (with the minION platform) compared to NGS technologies like Illumina. Finally, TGS offers increased portability and speed; minimal sample pre-processing is required, allowing smaller equipment sizes that can even collect and analyze the data in real-time. A downside to TGS is that read error rates are higher than for NGS (>10% against <2%)[4], which is why these technologies are sometimes used together in certain sequencing scenarios, which also makes up for their generally lower coverage than for NGS short reads.

On another note, Oxford Nanopore Technology (ONT) is at the heart of TGS technology and is one of the most commonly used techniques. ONT works by passing a single DNA molecule through a nanopore and measuring changes in the electrical field surrounding the pore as the molecule goes through, as shown in the video below:

[embed]How does Nanopore DNA Sequencing work? (Video Credits)

Despite its many benefits, difficulty in computational analysis of the sequencing procedure remains; the tools and algorithms developed are still seeing continuous improvement. As the technology gains popularity, users request more corrections and new features addressing the advances and novelties in the sequencing, and software that supports their original research projects. As already stated in their latest scientific article, published in Nature at the same time this blog was being written, ONT recognizes the need for:

“More robust and user-friendly bioinformatics software, such as cloud storage and computing and real-time analysis, will provide a further boost to ONT sequencing applications, ultimately moving the technology beyond the lab and into daily life.”

Megalodon

At the time of writing, there were 80 public repositories at ONT’s GitHub. These contain tools that can be used at multiple stages of ONT’s sequencing data processing and analysis. For a beginner, tackling the task of deciding which tool is the best for their needs and how to use it can be daunting. Among the pinned repositories, the **Megalodon suite **stands out for enabling several analyses in one single command. Relying on artificial neural networks, this algorithm is capable of producing the initial basecalls, reference mappings, modified basecalling, sequence variant calling, and more. In other words, all that is needed to begin processing ONT data.

Using the Megalodon suite presents its challenges. When running their analyses with Megalodon, users encountered some bottlenecks that indicated they needed additional support. These issues were the kind that could be insurmountable for small teams that do not have the access to powerful in-house hardware, a personalized-service cloud platform, and/or bioinformatics expertise. For example, the Megalodon tool itself makes use of the Guppy basecaller for performing the analysis. These applications require the use of GPUs for optimal performance, which is a not-so-common requirement in bioinformatic sequencing projects. Another challenge is efficiency can take a hit when processing large amounts of raw sequencing data. Even when providing generous computational resources to the analysis, users experienced that jobs in many occasions could slow down even to a halt. Jobs that had begun running at a processing speed of 90+ reads/s ended reducing their speed significantly as time progressed:

[2021/02/22 01:37:01] Read Processing: 99%|█████████▉| 3351851/3371560 [72:16:45<25:30, 12.88reads/s, samples/s=4.82e+5]

[2021/02/22 01:37:02] Read Processing: 99%|█████████▉| 3351857/3371560 [72:16:45<25:29, 12.88reads/s, samples/s=4.82e+5]

[2021/02/22 01:37:02] Read Processing: 99%|█████████▉| 3351863/3371560 [72:16:46<25:29, 12.88reads/s, samples/s=4.82e+5]

[2021/02/22 01:37:03] Read Processing: 99%|█████████▉| 3351869/3371560 [72:16:47<25:28, 12.88reads/s, samples/s=4.82e+5]

[2021/02/22 01:37:04] Read Processing: 99%|█████████▉| 3351875/3371560 [72:16:47<25:28, 12.88reads/s, samples/s=4.82e+5]

[2021/02/22 01:37:06] Read Processing: 99%|█████████▉| 3351881/3371560 [72:16:48<25:27, 12.88reads/s, samples/s=4.82e+5]

[2021/02/22 01:37:07] Read Processing: 99%|█████████▉| 3351887/3371560 [72:16:51<25:27, 12.88reads/s, samples/s=4.82e+5]

[2021/02/22 01:37:08] Read Processing: 99%|█████████▉| 3351893/3371560 [72:16:52<25:26, 12.88reads/s, samples/s=4.82e+5]

[2021/02/22 01:37:08] Read Processing: 99%|█████████▉| 3351899/3371560 [72:16:53<25:26, 12.88reads/s, samples/s=4.82e+5]

[2021/02/22 01:37:09] Read Processing: 99%|█████████▉| 3351905/3371560 [72:16:53<25:25, 12.88reads/s, samples/s=4.82e+5]

[2021/02/22 01:37:10] Read Processing: 99%|█████████▉| 3351911/3371560 [72:16:54<25:25, 12.88reads/s, samples/s=4.82e+5]

[2021/02/22 01:37:12] Read Processing: 99%|█████████▉| 3351917/3371560 [72:16:55<25:24, 12.88reads/s, samples/s=4.82e+5]

[2021/02/22 01:37:30] Read Processing: 99%|█████████▉| 3351923/3371560 [72:16:56<25:24, 12.88reads/s, samples/s=4.82e+5]

That was the last recorded speed of the runtime, as in fact all worker threads had disconnected and the job had been completely paralyzed:

2021–02–19 12:52:58.093083 [guppy/info] Client with id 19 has timed out.

2021–02–19 12:52:58.093134 [guppy/info] Client with id 19 has disconnected.

2021–02–19 12:53:03.093529 [guppy/info] Client with id 17 has timed out.

2021–02–19 12:53:03.093574 [guppy/info] Client with id 17 has disconnected.

2021–02–19 12:53:08.093956 [guppy/info] Client with id 20 has timed out.

2021–02–19 12:53:08.094014 [guppy/info] Client with id 20 has disconnected.

2021–02–20 18:37:04.418274 [guppy/info] Client with id 22 has timed out.

2021–02–20 18:37:04.420906 [guppy/info] Client with id 22 has disconnected.

2021–02–22 12:37:27.925837 [guppy/info] Client with id 21 has timed out.

2021–02–22 12:37:27.935710 [guppy/info] Client with id 21 has disconnected

Mind you, that problem apparently was fixed on version 2.3.0 of Megalodon, yet the way we overcame it provided additional benefits at runtime.

Nanopore tools in BatchX

This suite was implemented into the BatchX platform to bring out its full potential using the computational resources and expertise provided by BatchX. **BatchX offers a serverless infrastructure for running bioinformatics analyses as a fully managed service in the cloud**. Bioinformatics applications are implemented and made available so that users can apply them to their data while having complete control over how many computational resources they wish to assign to each job, as well as over other additional arguments they wish to include to modify their execution. In the context of this particular case, Megalodon was implemented in the platform (access here) so that users could have control of their testing needs and the possibility to optimize their workflow, being able to run their analyses with full transparency and as needed.

BatchX’s capacity to offer GPU resources not only tackled the GPU-need issue but also significantly increased the speed of the runs which in turn reduced the monetary cost of the analysis for the user. Generally, GPU resources are not made easily available from cloud providers and it falls upon the researchers to obtain their own GPUs, which can often set them back thousands of dollars and requires them to take care of the configuration and maintenance of new equipment whose investment may never be fully justified. Alternatively, running Megalodon on BatchX for one sample on the platform has an average computational cost of US$40 (for 150 GB of raw sequence data). Furthermore, a new pipeline was developed for resolving the issue of inefficient processing of large input data. The pipeline launches the same single Megalodon job in several smaller instances (batchx@bioinformatics/nanoporetech/megalodon-standalone) whose outputs are then merged, resulting in the same results being generated at one third or even smaller fraction of time and credit expense.

Megalodon is only the starting step when analyzing ONT data. After performing the initial processing with Megalodon, several analyses can follow depending on the researcher’s interest. Examples include performing the same analysis using the **Nanopolish software package, calling copy number variations with [CNVkit](https://cnvkit.readthedocs.io/en/stable/), structural variations and variant phasing with [Sniffles](https://github.com/fritzsedlazeck/Sniffles)**, among other possibilities. In the same manner as was done for simplifying the user’s workflow with Megalodon, for automating the posterior analysis of a sequencing run the analysis tools were wrapped into pipelines that would execute the whole workflow in a single command (e.g., batchx@bioinformatics/nanopore/sv-pipeline). This is one of the pipelines for Nanopore analysis, specifically tailored for the identification of structural variants.

Directed Acyclic Graph (DAG) displaying the workflow executed in an analysis of ONT data on the BatchX platform using the pipeline batchx@bioinformatics/nanopore/sv-pipeline.

Oxford Nanopore offers an amazing and revolutionary technology; it doubles down on the improvements brought by the NGS wave introducing numerous benefits. However, Nanopore analysis can be a daunting task for researchers first attempting to perform their own pipelines. And even for laboratories with established protocols, the long analysis times, and the need for GPUs, efficiency, and reproducibility between runs present a challenge that is normally disregarded due to heavy workloads, in spite of its importance. The complex analysis required to process the TGS sequencing data raises the need for powerful computation infrastructure and tailored bioinformatic algorithms that are not so achievable for every lab or research group. BatchX provides the solution you have been looking for, providing ready-to-use bioinformatics solutions, infrastructure, and personalized support so that any institution can set off working with this outstanding technology in the blink of an eye, welcome to the revolution!

A quote from one of our users:

From the genomics lab at VHIO, we wanted to describe the methylome profile of a specific cell line under different conditions.

We wanted to use TGS because long reads were going to give us information regarding long structural variants and also CpG distribution throughout the genome.

We decided to use Oxford Nanopore technologies because the technology, the kits, and the sequencer were very accessible and cost-efficient.

We tried to set up the downstream analysis after the basecalling with Guppy but we didn’t have the computer power required and the data processing was going very slowly.

We contacted BatchX to ask if they could help us and they not only provided the necessary computer power but also the pipeline ready for a non-bioinformatic user.

BatchX helped us to save an immense amount of time and money. Their pipelines and platform are very user-friendly and they were always available if I had any question or doubt.

For research groups wanting to start sequencing with little funding or little bioinformatic knowledge, I highly recommend BatchX as a platform to process and analyze your sequencing data.

Happy sequencing 😊

**- Paula Romero Lozano, **NGS Developer Scientist at Vall d´Hebron Institute of Oncology (VHIO)

A special thanks to Marcus Stoiber for being so attentive in responding to our issues on GitHub.

1. https://emea.illumina.com/science/technology/next-generation-sequencing/plan-experiments/read-length.html
2. Heather, J. M., & Chain, B. (2016). The sequence of sequencers: The history of sequencing DNA. Genomics, 107(1), 1–8. https://doi.org/10.1016/j.ygeno.2015.11.003
3. https://nanoporetech.com/
4. Lu, H., Giordano, F., & Ning, Z. (2016). Oxford Nanopore MinION Sequencing and Genome Assembly. Genomics, Proteomics and Bioinformatics, 14(5), 265–279. https://doi.org/10.1016/j.gpb.2016.05.004