Consider working on genomics

(chiming in here as a founding engineer at Atomic)

So I spent more than 8 years as a SWE at Google, and now work here with both experimental biologists and machine learning scientists. And yes, a lot of the concerns mentioned in this thread are also things I have had anxiety about.

Most obvious to me, being a software engineer at Google felt like being the center of the universe. Coming here, the focus is the scientific research. And yes, the scientists all managed to complete their PhDs so they don't necessarily need me to unblock them every second of their day. But contrary to my expectations, this has been remarkably freeing. I think one particularly important part of our company that makes this work is that, even on the science side, we're multidisciplinary (at a high level, emphasizing both experimental biology and ML.) And so engineering feeling like another arm of that multi-discipline nature is fairly... natural.

The reason I feel it's freeing, and the reason I enjoy working here, is also the greatest challenge. Because the scientists are focused on the science, because they respect me and trust me to figure it out, and because they aren't constantly blocked by me, my job is mostly about dreaming extremely expansively about what I can do to reduce toil and make the scientists more productive. Of course they have feedback and input, but how I use my time and what I build is ultimately my decision because I am the engineer. And I have been able to do some things I am very proud of, like rolling out Bazel and Kubernetes and finding ways to seamlessly bring them into the cloud (we're even multi-cloud now without them even noticing!) On the other hand, it's very challenging because when you work on a product, say Google Photos, as a SWE, you always have some direct tether to the product ("what should we build next? ahhhh, well I guess we could just embed stable difficusion and a million people would immediately play with it".) At Atomic, my tether is very ambiguous. If I do my job successfully, they'll be able to do research more quickly (? effectively?), and eventually we'll be able to produce a therapeutic that hopefully changes the world. Identifying what I can do today to speed up that far outcome in the future is very challenging, but it is a far more interesting challenge than gluing some pre-existing software into my UI or running A/B tests to turn a red button blue.

If, like me, you enjoy being given ownership over incredibly ambiguous problems, please do reach out!

This role focuses on directly partnering with the biologists: https://boards.greenhouse.io/atomai/jobs/4726839004

This role is expansive cloud infra: https://boards.greenhouse.io/atomai/jobs/4531035004

And this role is directly partnering with the ML scientists: https://boards.greenhouse.io/atomai/jobs/4191285004

I'd like to hear about this too!

As someone who puts tremendous value in technical mentorship when considering a role this is about the worst possible advertisement for being a swe in genomics as it amounts to "all our code is awful- come fix it!"

Protein structure prediction was a huge deal, which is why AlphaFold received so much fanfare. It is actually pretty good. The next step is to predict where multi-protein complexes would interact- which is not just as simple as predicting the structure of two proteins independently and then trying to fit them together like a puzzle, because the the interactions can also change the structure. While it's not as hard as it used to be to experimentally determine protein targets of, for example, a protein kinase, it's still not an arbitrary or cheap experiment, and to do that for the many thousands of such proteins, across different conditions (stress, presence of co-factors, etc) and in different organisms would be rather a lot of work. Something like alphafold that makes reasonable predictions and can be used to help you focus on what's most likely to be relevant to your disease or process of interest helps quite a bit.

There's also more need for integrating "multi-omics" data, where you have data from multiple assays (gene expression, phospho-proteomics, lipidomics, epigenetics, small RNA expression, etc etc) with the goal of somehow combining all these different assay results from various levels of gene regulation, to get closer to figuring out actual mechanism for complex processes. Building on that, we can also do single-cell multi-omics to some extent- where you have results from different sequencing-based assays on the level of the same individual cell. This is still pretty limited, but it's exciting and advancing pretty quickly. This will eventually be combined with things like spatial transcriptomics, which is useful for mapping out what's going on in heterogeneous tissue samples like tumors, for example, so we'll end up with spatial single-cell multi-omics, at which point you're looking at 1) some quantitative trait for multiple genes/loci/molecules, and often 10k+ of such features at the same time per assay, 2) multiple assays, such as DNA accessibility and gene expression, in 3) single-cells, of which you might have 10k of in a single sample, 4) across a physical tissue sample where individual cells are spatially mapped, and where you probably want to figure out how cells might influence the state of those around them, and 5) in multiple different samples, where you might want to compare disease vs control, or look for correlation to heterogeneity of results within one group.

There's a lot of public data already available for single-cell gene expression projects if you want to get a feel for how these things are structured and how (passable but not amazing) the existing tooling is- one of the main repositories for this data is the NCBI's SRA https://www.ncbi.nlm.nih.gov/sra but you'll quickly note that searching and browsing is not as easy as you might think it would be- because one of the main limiting factors in bioinformatics is how bad everyone is at keeping terminology consistent. For many bioinformaticians, a majority of time is spent in the data cleaning phase. It's awful. Sometimes the experimental parameters make it into SRA or GEO, but sometimes you have to read through the associated paper to pull that out. Often it's only large consortium projects like the The Cancer Genome Atlas (TCGA) or the Genotype-Tissue Expression project (GTEx) - which have enough funding for staff dedicated to data management- end up publishing datasets that are easy to "consume" without having to jump through a whole bunch of hurdles to figure out how the data was produced.

I have a BS/MS in bioinformatics and I'm presently a PhD candidate in genetics and computational biology defending in February.

Invent a new file format (or a few) for storing genomics data. They're all the rage in the bioinformatics field. Make sure not to document its semantics so that its implementation is the only spec.