NAF: Text vs DNA mode

The NAF compressor has "--dna" mode for DNA sequences, and "--text" mode that supports any text sequences. Since "--text" can be used for DNA data as well, the question is - which mode should be used for DNA sequences? This page briefly explores this question.

Theoretical considerations

• For compression time, in "--dna" mode, the sequences are first converted into 4-bit representation, which is then fed into zstd compressor. Since zstd now deals with twice smaller data, the compression should occur twice faster (which is the rationale for using 4-bit presentation in the first place). And the process of converting into 4-bit is so fast that its time can be ignored here.
• As for compactness, the difference should be small at stronger compression levels. However with low (fast) compression levels, "--dna" mode should provide measurable advantage.

Benchmark

I benchmarked ennaf settings from "-1" to "-22" with "--dna" and "--text" on a test genome.

• Test dataset: Strongylocentrotus purpuratus (sea urchin) genome, RefSeq accession GCF_000002235.4.
  • Size of FASTA file: 1,007,867,539 bytes
  • Number of sequences: 31,897

Results

Interpretation

• "--dna" mode speed (for both compression and decompression) is clearly better than "--text" mode at extreme levels ("-1" and "-22").
• For compactness, "--dna" mode is much better than "--text" at "-1" level. At "-22" level the difference is small.

Other datasets

I tried a number of other datasets (data not shown here), and they show similar overall pattern. Especially the speed difference is very similar to the above results.

As for compactness, at "-1" level, "--dna" mode is always much better. At "-22" the two modes are close, and sometimes "--text" is more compact. On average, with the datasets I tried so far, "--dna" has stronger compression even at "-22" level.

What about the other modes?

The "--rna" mode performs identically with "--dna" (the only difference between them is the accepted sequence alphabet), and "--protein" mode is just a more restrictive version of "--text". So these results apply to "--rna" and "--protein" modes as well.

One thing to note is that with protein sequences, there is no performance (size or speed) difference between "--protein" and "--text" modes. The only reason to use "--protein" mode is for stricter protein-specific validation of the alphabet used in the input.

Conclusion

"--dna" mode should be used for DNA sequences whenever possible, as it's faster and more compact. Similarly, "--rna" should be used with RNA data. For protein data, "--protein" and "--text" modes don't differ in speed or compactness, but only in how they validate the input.

Of course, if your data includes non-standard characters, you may have to use "--text" mode regardless, as preserving the data intact is usually more important than any performance gain.