Feature: Support LJA for genome size estimation in helper command

[MrTomRod]

Description

This PR modifies the genome_size helper command to support LJA as an alternative assembler. Previously, the command hardcoded the use of Raven.

Motivation: LJA is significantly faster than Raven (though it may only be used for PacBio HiFi data?)

Usage Example:

# Default behavior (uses Raven)
autocycler helper genome_size --reads reads.fastq.gz

# Use LJA (faster for HiFi)
autocycler helper genome_size --reads reads.fastq.gz --assembler lja

Performance

Benchmarking on a test dataset shows LJA is approximately 6x faster in wall-clock time and 9x faster in CPU time compared to Raven:

Assembler	Real Time	User Time	Sys Time
LJA	1m39.646s	9m24.194s	0m6.220s
Raven	10m39.524s	82m46.248s	0m14.592s

Note: LJA support is currently limited to PacBio HiFi reads.

Changes

• CLI: Added --assembler argument to autocycler helper genome_size.
• Logic: Updated src/helper.rs to handle the new argument.
  • If --assembler lja is passed, it calls genome_size_lja.
  • If --assembler raven (or nothing) is passed, it defaults to genome_size_raven.

[rrwick]

Hi Thomas,

Thanks for this! Some thoughts/observations:

• Even though LJA is advertised as a HiFi assembler, it can be used for ONT reads, though its speed performance suffers (see Figure S5 from the Autocycler paper). So Raven is probably preferable for ONT reads.
• I tried LJA, Raven and LRGE on an E. coli HiFi read set (same one from this post, first 40k reads):
  • Using 8 threads, LJA took 1:55, Raven took 2:30, LRGE took 0:28.
  • Using 32 threads, LJA took 1:21, Raven took 0:52, LRGE took 0:10.
  • My LJA assembly was very accurate for genome size estimation (only off by 1 bp). But the Raven assembly size was definitely good enough (0.15% error), since Autocycler only needs an approximate genome size. LRGE had the most error (~5%), but this is still good enough for Autocycler.

I'm curious why you got much better speed performance with LJA than I did. What was your read set like? How deep? How long were the reads? How big was the genome?

My inclination is to not merge this PR, since it adds a bit of complexity to the tool and I'm not sure it's needed. I think Raven is a good default choice (often faster than LJA in my tests), and If users really need faster genome size estimation, they can use LRGE. But am I missing something?

Thanks,
Ryan

[MrTomRod]

I didn't realize that LJA is sometimes slower than Raven. I never observed this, but I'm always using PacBio HiFi reads.

Barcode Quality	HiFi Reads	HiFi Read Length (mean, bp)	HiFi Read Quality (median, QV)	HiFi Yield (bp)	Polymerase Read Length (mean, bp)	FASTQ Size	Genome Size	Coverage
98.4	134,289	7,783	Q47	1,045,225,151	152,594	430 MB	2.5 Mbp	~400x

Feel free to reject the PR.

I also tried LRGE and am using that now, too.

Happy holidays!