DIAMOND v2.1.0

[bbuchfink]

• Added the cluster workflow to cluster protein sequences.
• Added the realign workflow to generate clustering output.
• Added the recluster workflow to correct errors in clusterings.
• Added the reassign workflow to reassign cluster members to their closest centroid.
• Added the option -M/--memory-limit to set a memory limit for clustering workflows.
• Added the approx-id option to filter alignments by approximate sequence identity and to set an approximate sequence identity threshold for clustering.
• Added the --member-cover option to set the coverage threshold of the cluster member sequence.
• Added the cluster-steps option to set steps for cascaded clustering.
• Added the clusters option to specify clustering input file.
• The blastx mode will now mask any open reading frame below the minimum required length as specified by --min-orf.
• The blastx mode will only count unmasked letters towards the block size.
• Fixed a bug that caused an error when using the global ranking mode.
• Added the fast mode as the first round in iterative searches.
• Fixed a bug that caused the program not to function on systems without support for SSE4.1.
• Improved multi-threaded load balancing of gapped extension computations.
• Improved performance of seed extension stage when HSP filter settings are used.
• Added the option --soft-masking with possible values 0 and tantan to permit soft-masking using the tantan algorithm.
• Fixed a bug that could cause an inflate error in multiprocessing mode.
• Added the option --swipe to compute full Smith Waterman alignments of all queries against all targets.
• Added the sensitivity mode --faster.
• Added the option -M to set memory limit for the cluster, realign, recluster and reassign workflows.
• Added the output fields approx_pident and corrected_bitscore to the tabular format.
• Added the --lin-stage1 option to linearize comparisons in the seeding stage by only considering hits against the longest query sequence for identical seeds.
• Added the --kmer-ranking option to rank sequences when --lin-stage1 is used (only supported when compiled with -DKEEP_TARGET_ID=ON).
• Added the option --no-block-size-limit to deactivate upper limits for the block size when the --memory-limit option is used.
• Added the greedy-vertex-cover workflow to compute clustering based on alignments.
• Added the --edge-format option to set edge format for greedy vertex cover.
• Added the --edges option to set input file for greedy vertex cover.
• Added the --centroid-out option to output centroid sequences for greedy vertex cover.
• Added the --unaligned-targets option to generate an output file of unaligned targets.
• Fixed an issue that failed compilation using the Intel Compiler.
• Fixed an issue that could cause a segmentation fault in rare cases.
• The --header option can now be used with the parameter simple to enable simple headers for the tabular format, or without a parameter to enable headers for the clustering format.
• Added the option --mp-self to optimize self-alignment in multiprocessing mode.
• Added the option --query-or-subject-cover to report alignments if the query or the subject cover (or both) are above the given threshold.
• Removed support for the --comp-based-stats 2 option (now equivalent to --comp-based-stats 3).
• Removed hit culling in case of overlapping target ranges in frameshift alignment mode.
• Added the option --anchored-swipe to activate anchored SWIPE extension.

---

This discussion was created from the release DIAMOND v2.1.0.

[taylorreiter]

Would you be willing to provide an explanation for the corrected_bitscore? I tried to search for documentation on what type of correction this is, but I couldn't find it. I was very excited to see this addition though!

[bbuchfink]

It's basically the correction described here: https://bmcresnotes.biomedcentral.com/articles/10.1186/1756-0500-5-286

I adapted this for correcting the bit score instead of the e-value (based on the output of the ALP library).

[taylorreiter]

Thank you so much for your response, this is super helpful to see!

[V-Varga]

I apologize for the trouble, but could you elaborate on how you calculate percent identity for --approx-id? I have searched the documentation, but the best I can find is "the approximate sequence identity derived as a linear regression from the bitscore, not the actual number of identities in the alignment" from the Wiki page for Clustering (https://github.com/bbuchfink/diamond/wiki/Clustering). This sounds distinct from the BLAST definition (as I understand it: the number of identical residues/amino acids divided by the number of alignment columns).

[bbuchfink]

The formula to compute it is given in the supplementary information of our clustering preprint.

[V-Varga]

Thank you very much for taking the time to respond! This was very helpful, exactly what I was looking for!