if your instance is not hibernated you will be paying for it, regardless of mamba. you will pay for the EBS volume even if you hibernate.

there are frameworks like ParallelCluster which recreate clusters in AWS, but assuming you are on a framework like Nextflow it is usually preferable to use AWS Batch. There is also the Amazon Genomics CLI. They have a lot of information about this https://aws.amazon.com/health/genomics/solutions/. Biostars is more suitable if you have a specific question.

The ENCODE consortium runs workshops that include hands-on sessions to work with the data and pipelines, including AWS accounts as needed. Larger sessions are also under the "ENCODE Jamboree" name, in case that is useful for getting more info from the consortium. That might provide some flavor on the analysis side of things.

Amazon recently held a talk on "Scaling genomics workloads using HPC on AWS". Part of that talk went into how to integrate existing Nextflow and other pipelines to transition from a local HPC to AWS-based cloud work. It might be worth poking around to find this and other talks. There's another talk on "Automating Genomics Workflows on AWS" that might be relevant, for instance.

Other parts of that particular talk on scaling went into more nuts-and-bolts stuff about using spot instances and cold storage to lower overall costs, as AWS is expensive. In the case of spot instances, you bid on a machine instance. If your bid is accepted, you apply a pre-built image to that instance and then get your pipeline running. If someone outbids you, Amazon can "interrupt" your host and you may lose it, so your image and pipeline should be lean, fast, and fault tolerant. If you use a dedicated host, however, then you will pay more, but you won't have to bring it down and reimage on reboot, or worry about it getting interrupted — it stays persistent until you terminate it.

You can use images so that every time, you don't have to set up tools.

Can you use something like Nextflow executed on AWS / Biocontainers? You get a lot of of bioconda tools you can chain, and that should be portable.

In case you weren't aware, bioinformatics platform technology has moved on quite a bit in recent years so I thought I would point out an alternative approach for those in the same situation. If you don't want to spend all the time and resources to set things up yourself in AWS, then second generation bioinformatics platforms like Basepair (and LifeBit I believe) enable you to simply plug in the EC2 and S3 resources in your own AWS account after you have created it. Basically the platform then abstracts away all of the devops that you would normally have to do yourself, yet acts as an easy wrapper around industry standard tools and pipelines for a multitude of NGS data types. You can even deploy and run your own workflows if you'd prefer, and with a per sample usage model that includes 12 months of storage, it's pretty economical if you consider all of the time you save.

Great question — AWS is powerful but the billing complexity can be brutal for bioinformatics teams. A few things that helped us and teams we work with:

1. Spot instances are non-negotiable. For Nextflow or any batch genomics workload, spot can cut costs 60-80%. The key is having automatic retry logic so interrupted jobs just re-queue.

2. Scale to zero matters more than you think. Most genomics teams run bursty workloads — a big WGS batch Monday, nothing Tuesday. If your cluster doesn't scale to zero between runs, you're paying for idle compute 24/7.

3. Avoid over-provisioning instance types. Tools like Nextflow's resource labels help, but you need the compute layer to actually right-size. AWS Batch does okay here but has significant overhead per job.

We built Clusterra specifically for this use case — managed Slurm on Kubernetes with native spot handling, scale-to-zero via Karpenter, and pass-through spot pricing ($0.0173/vCPU-hour). Happy to share more details if useful, but the spot + scale-to-zero combo is the biggest cost lever regardless of what platform you use.

Disclosure: I'm the founder of Clusterra.