WIP: adjacency-aware tile filtering

[danilexn]

(e.g., for Open-ST data)

This implements a new, more 'sophisticated' strategy for filtering tiles.

On top of thresholding the % of barcode matches per tile, this adds the spatial (physical) adjacency structure of the tiles/pucks, detects connected subgraphs with weighted partitioning, keeps the subgraph with the highest % matching, and lastly applies a "dilation" strategy to also include tiles surrounding the ones that passed filters so far.

It includes helper scripts to generate the "adjacency" graph for the NovaSeq 6000 S4; it generates a networkx edgelist file, which provides more flexibility for adapting to other FCs/types of spatial puck. The script for computing spatial matches implements all functions for graph partitioning and tile "dilation".

Also, I updated the coordinate system csv file to add two new columns (lane and side). This does not have any conflict with the previous file structure. We can consider adding more columns to reduce the dependency on the tile/puck IDs for parsing the tile column/row.

To-Do:

• Implement this into the snakemake workflow
• Write unit tests
• Maybe add more columns to coordinate system files