Bioinformatics & Computational Single-Cell Genomics Expert

$70–100/hrrate as listed

Design graduate-level single-cell genomics problems that test whether frontier AI models can run real analysis pipelines in scanpy, scvelo, squidpy, and gudhi. You write the problem setup, the oracle function, and the validator, then iterate against model attempts until the difficulty lands where the benchmark needs it.

Summary, eligibility guidance and screening preparation are prepared by Sidequest from the platform's posting. Check the source listing before applying.

The work

You are a task designer for a scientific reasoning benchmark, not an annotator. Each task starts from a real single-cell or spatial workflow you have actually run — QC and normalization choices, cell-type annotation, pseudotime ordering, RNA velocity, multi-omic integration, spatially variable gene detection, persistence-based analysis — and gets converted into a problem with a verifiable answer. Two flavors dominate: fully specified problems where the model must execute a long multi-step pipeline correctly, and partial-information problems where the model must plan a sequence of queries or in-silico experiments to recover something the data does not show directly.

After designing, you test. Every problem runs against current frontier models, and you tune it until it sits in the target band — hard enough that pattern matching fails, clean enough that a competent postdoc with the same tools would agree on the answer. Expect to throw work away when a model solves it on the first try, or when your own oracle turns out to be ambiguous under a second reading.

What the screen looks for

Real tool mileage. Which scanpy defaults you have overridden and why; where scvelo's dynamical model misbehaves; what squidpy's neighborhood tests assume; how you set gudhi filtration parameters on gene expression data.
Verifiable provenance. Publications, preprints, open-source commits, or professional pipelines you can point to and discuss under follow-up.
Python engineering. You will write setup scripts, deterministic oracle functions, and validators that tolerate legitimate numerical variation without accepting wrong answers.
Problem-design instinct. The ability to distinguish genuinely hard from merely tedious, and to spot the shortcut that lets a model guess your answer.

Logistics

Fully remote and asynchronous, with work done in Linux remote compute sandboxes; containerized, reproducible setups are a plus. The listing asks for 15–20 hours per week minimum, and throughput matters more than fixed hours — this is described as the platform's highest-volume domain, so accepted contributors often see steady task flow. Observed pay for this listing is $70–100/hr, set by the platform based on screening outcome and domain depth; it is not a guarantee.

What they look for

Graduate-level training (MS/PhD or equivalent research experience) in bioinformatics, computational biology, or a related STEM field
Hands-on production use of scanpy, scvelo, squidpy, or gudhi on real research data
Strong Python — writing problem setups, deterministic oracle functions, and solution validators
Depth in single-cell and spatial methods: cell-type annotation, pseudotime/trajectory inference, multi-omic integration, spatially variable genes, persistent homology
Comfort in Linux/terminal environments and remote compute sandboxes
Demonstrable output: publications, preprints, or open-source contributions using these libraries

Before you apply

Would you pass this screen?

Drawn from what this listing actually requires. Answered in your browser — nothing is sent to us, and we do not store your answers.

Do you hold a Master's degree or PhD in a STEM field, or have equivalent research experience (e.g., 5+ years as a research scientist or engineer in a lab or biotech role)?
Graduate STEM training · required
Have you used scanpy, scvelo, squidpy, gudhi, or another specialized bioinformatics library to analyze real data in a research project, publication, or professional role?
Hands-on single-cell or bioinformatics software proficiency · required
Have you written Python code (beyond interactive scripts or notebooks) for data analysis, validation, or scientific workflow automation, as shown in published work, open-source contributions, or professional projects?
Python for scientific computing · required
Are you comfortable working in a Linux terminal and have experience with remote compute environments, version control (git), and package management (conda/pip/containers)?
Linux/terminal proficiency · preferred
Can you commit to at least 15–20 hours per week for this role over the duration of the project, with flexibility to adjust based on testing cycles?
Weekly availability · required

0 of 5 answered

Included with this listing

Interview prep — 11 questions

Matched to what this platform's screening actually measures. Answer each out loud in 60–90 seconds: claim → concrete example → implication.

What it tests: Whether you've actually hands-on built workflows with these libraries to answer real questions, not just run tutorials; whether you understand the tool's design choices and when it's the right fit versus an alternative.
A strong answer: Name a specific project (publication, thesis work, or professional analysis), identify the biological goal clearly, explain why you picked that tool over others, walk through a key technical decision you made, and describe what you found or learned that validated the approach.
Where people lose it: Describing a tool's features in general terms ('scanpy is great for clustering') without tying it to a concrete problem you solved, or glossing over why you chose it instead of saying 'my advisor told me to use it.'