Performance of GO enrich

[slokmanex]

Hi,
Your tool is excellent, however, over representation analysis is very slow for GO compare to other terms (KEGG, DOSE). I think it might have to do with the amount of GO data.

I hope the performance improves in the future.

[GuangchuangYu]

1. GO is larger than KEGG & DO.
2. extra works need to do for processing the OrgDb object to prepare annotation suitable for analysis (this step can be optimized).

[RestlessTail]

Hi @GuangchuangYu,

Thank you for maintaining this incredibly useful package over the years. I think a two-level cache may help. Here’s a quick diagram of the idea:

graph TD
    A[First run] -->|Build mappings| B[L1: Full goAnno cache]
    B -->|Split by ontology| C[L2: GOdb$BP/CC/MF/ALL]
    D[Subsequent calls] -->|Read from| C
    C -->|If missing| B

Loading

• L1 Cache: Stores raw GO-gene mappings (goAnno)
• L2 Cache: Ontology-specific environments
  • BP/CC/MF: Pre-built ontology db
  • ALL: Lightweight environment (only GO2ONT)

I've implemented a modified version: https://gist.github.com/RestlessTail/6dcf93eea79ad7918738d53b8e1c6c80.

Metric	Original	Cached	Improvement
Time (mean)	18.7s	3.2s	5.8x faster
Memory	71MB	117MB	+46MB

The modified version keeps the external API unchanged, while reducing compute time significantly for repeated calls. Full benchmark script here: https://gist.github.com/RestlessTail/36a9ce238e25d82177d3ff21a704128d. Would this optimization be useful to integrate?