gwtaylor
commented
1 month ago Hi @hjarnek, thank you for your interest in PROTAX-GPU and for reaching out!
Documentation Updates
We acknowledge that our current documentation is lacking, and we're actively working on improving it. The lead developer is in the process of transitioning dependencies from the Chex library to Flax to better integrate with the JAX ecosystem. This change aims to streamline future development and enhance performance. We delayed updates to the documentation during this transition but expect to release comprehensive instructions soon, irrespective of our progress on Flax.
Translation and Alignment Steps
Regarding your question about the translation and alignment steps mentioned in our paper:
"[...] before computing sequence distances, following [13], the query and reference sequences are first translated into amino acid sequences, then aligned using a hidden Markov model (HMM), and finally back-translated into nucleotide sequences."
Purpose of Translation and Back-Translation
This approach is commonly used for protein-coding genes like the cytochrome c oxidase I (COI) gene. Here's why:
- Conservation of Amino Acid Sequences: Amino acid sequences are more conserved than nucleotide sequences due to the redundancy of the genetic code. This means that aligning amino acid sequences can be more reliable.
- Improved Alignment Accuracy: Translating nucleotide sequences into amino acids allows for better alignment using HMMs tailored for protein sequences.
- Back-Translation: After alignment, the sequences are back-translated to nucleotides to retain the original genetic information for downstream analyses, like PROTAX-GPU.
You can find more details on this method on pages 42 and 44 of this guide.
Applicability to Non-Protein-Coding Genes
For non-protein-coding genes like rRNA genes or the Internal Transcribed Spacer (ITS) regions:
- No Translation Step: Since these genes do not code for proteins, the concept of translating to amino acids doesn't apply.
- Direct Nucleotide Alignment: These sequences are typically aligned directly at the nucleotide level using tools designed for non-coding DNA.
- HMM Alignment: While HMMs can still be used for alignment, they are specifically designed for nucleotide sequences in this context.
Genetic Translation Tables
Regarding different genetic translation tables:
- Customization Needed: The translation step would indeed need to be adapted if you're working with organisms that use alternative genetic codes.
- Software Configuration: Alignment tools often allow you to specify the genetic code, ensuring accurate translation for various organisms.
Alignment and PROTAX-GPU
It's important to note:
- Alignment Outside the Codebase: The alignment code we talked about above is not in our codebase. PROTAX-GPU assumes that your sequences are already aligned. The software focuses on taxonomic assignment rather than sequence alignment.
- Preprocessing Required: You'll need to perform alignment using appropriate tools (e.g., MAFFT, MUSCLE) before inputting your data into PROTAX-GPU.
hjarnek
Hi,
This looks super-cool, but do you plan to publish any instructions on how to use it also? :)
Another small question. You state in your paper for PROTAX-GPU that...
What's the idea with this step? How does that work with non-protein-coding genes, like rRNA genes, and how does it account for different genetic translation tables?
Cheers