Handle very large datasets (>100M single cells)

[shntnu]

Come up with a solution for ingesting CSVs into a backend that is more scalable than SQLite.

Context: The package cytominer-database is a key component of image-based profiling workflows: it is a small Python-based command line tool that ingests data generated from CellProfiler into a database. Currently, only SQLite is supported but it would be great if we could use something more scalable. Addressing this issue would equip researchers working with single cell imaging data to execute queries and perform analysis across tens of millions of cells. This would be particularly useful to analyze single cell data across all plates in an experiments, or across multiple experiments.

E.g. a recent 135 plate experiment had 100M single cells and there's no easy way to analyze this.

Here's how to started on this issue:

• Read the ingest documentation here
• Test out cytominer-database by running the tests yourself
• Read the code
• Now focus on this part of the code, where the ingestion into the backend happens
• Come up with ideas on how this could be done at scale and discuss it here
• if you have any questions, please use this issue to discuss or open a new issue

[gwaybio]

I would just add that if there any questions, to add them here or open a new issue - intro steps look great 👍

[shntnu]

https://parquet.apache.org/ looks like a good option for this