From p.j.a.cock at googlemail.com  Tue May  3 05:24:08 2011
From: p.j.a.cock at googlemail.com (Peter Cock)
Date: Tue, 3 May 2011 10:24:08 +0100
Subject: [Open-bio-l] Interesting BLAST 2.2.25+ XML behaviour
In-Reply-To: <BANLkTi=6_2bFpGhOwxtdjy-DzxUotVWxEg@mail.gmail.com>
References: <BANLkTi=6_2bFpGhOwxtdjy-DzxUotVWxEg@mail.gmail.com>
Message-ID: <BANLkTinH0y4KQ7_AXt7Ly3TgN9fXxErUzA@mail.gmail.com>

Hello all,

I've CC'd the BioPerl, BioRuby, BioJava and Biopython development mailing
lists to make sure you're aware of this, but can we continue any discussion
on the cross-project open-bio-l mailing list please?

I noticed that recent versions of BLAST are not using a single <iteration>
block for each query, which was the historical behaviour and assumed
by the Biopython BLAST XML parser. This may be a bug in BLAST.
See link below for an example.

Has anyone else noticed this, and has it been reported to the NCBI yet?

Thanks,

Peter

(Not for the first time, I wish there was a public bug tracker for BLAST,
or at least a private bug tracker so we could talk about issues with an
NCBI assigned reference number.)

---------- Forwarded message ----------
From: Peter Cock <p.j.a.cock at googlemail.com>
Date: Wed, Apr 20, 2011 at 6:08 PM
Subject: Interesting BLAST 2.2.25+ XML behaviour
To: Biopython-Dev Mailing List <biopython-dev at biopython.org>


Hi all,

Have a look at this XML file from a FASTA vs FASTA search
using blastp from ?BLAST 2.2.25+ (current release), which
is a test file I created for the BLAST+ wrappers in Galaxy:

https://bitbucket.org/galaxy/galaxy-central/src/8eaf07a46623/test-data/blastp_four_human_vs_rhodopsin.xml

I just put it though the Biopython BLAST XML parser, and
was surprised not to get four records back (since as you
might guess from the filename, there were four queries).

It appears this version of BLAST+ is incrementing the
iteration counter for each match... or something like that.

Has anyone else noticed this? I wonder if it is accidental...

Peter


From cjfields at illinois.edu  Tue May  3 09:31:55 2011
From: cjfields at illinois.edu (Chris Fields)
Date: Tue, 3 May 2011 08:31:55 -0500
Subject: [Open-bio-l] [BioRuby] Interesting BLAST 2.2.25+ XML behaviour
In-Reply-To: <BANLkTinH0y4KQ7_AXt7Ly3TgN9fXxErUzA@mail.gmail.com>
References: <BANLkTi=6_2bFpGhOwxtdjy-DzxUotVWxEg@mail.gmail.com>
	<BANLkTinH0y4KQ7_AXt7Ly3TgN9fXxErUzA@mail.gmail.com>
Message-ID: <398303E2-1195-4CC2-8B73-09C6C1117892@illinois.edu>

Haven't tried this using the latest BLAST+ myself, but it doesn't surprise me too much.  Also agree re: some kind of bug tracking with NCBI; I believe they have an internal one, but it would be nice to have a public interface to it.

chris

On May 3, 2011, at 4:24 AM, Peter Cock wrote:

> Hello all,
> 
> I've CC'd the BioPerl, BioRuby, BioJava and Biopython development mailing
> lists to make sure you're aware of this, but can we continue any discussion
> on the cross-project open-bio-l mailing list please?
> 
> I noticed that recent versions of BLAST are not using a single <iteration>
> block for each query, which was the historical behaviour and assumed
> by the Biopython BLAST XML parser. This may be a bug in BLAST.
> See link below for an example.
> 
> Has anyone else noticed this, and has it been reported to the NCBI yet?
> 
> Thanks,
> 
> Peter
> 
> (Not for the first time, I wish there was a public bug tracker for BLAST,
> or at least a private bug tracker so we could talk about issues with an
> NCBI assigned reference number.)
> 
> ---------- Forwarded message ----------
> From: Peter Cock <p.j.a.cock at googlemail.com>
> Date: Wed, Apr 20, 2011 at 6:08 PM
> Subject: Interesting BLAST 2.2.25+ XML behaviour
> To: Biopython-Dev Mailing List <biopython-dev at biopython.org>
> 
> 
> Hi all,
> 
> Have a look at this XML file from a FASTA vs FASTA search
> using blastp from  BLAST 2.2.25+ (current release), which
> is a test file I created for the BLAST+ wrappers in Galaxy:
> 
> https://bitbucket.org/galaxy/galaxy-central/src/8eaf07a46623/test-data/blastp_four_human_vs_rhodopsin.xml
> 
> I just put it though the Biopython BLAST XML parser, and
> was surprised not to get four records back (since as you
> might guess from the filename, there were four queries).
> 
> It appears this version of BLAST+ is incrementing the
> iteration counter for each match... or something like that.
> 
> Has anyone else noticed this? I wonder if it is accidental...
> 
> Peter
> 
> _______________________________________________
> BioRuby Project - http://www.bioruby.org/
> BioRuby mailing list
> BioRuby at lists.open-bio.org
> http://lists.open-bio.org/mailman/listinfo/bioruby



From p.j.a.cock at googlemail.com  Wed May  4 06:36:57 2011
From: p.j.a.cock at googlemail.com (Peter Cock)
Date: Wed, 4 May 2011 11:36:57 +0100
Subject: [Open-bio-l] [BioRuby] Interesting BLAST 2.2.25+ XML behaviour
In-Reply-To: <4DC12371.3040204@gmail.com>
References: <BANLkTi=6_2bFpGhOwxtdjy-DzxUotVWxEg@mail.gmail.com>
	<BANLkTinH0y4KQ7_AXt7Ly3TgN9fXxErUzA@mail.gmail.com>
	<398303E2-1195-4CC2-8B73-09C6C1117892@illinois.edu>
	<4DC12371.3040204@gmail.com>
Message-ID: <BANLkTinV4uha74Y9jC_f=XLK5LufAn1xHw@mail.gmail.com>

On Wed, May 4, 2011 at 10:59 AM, Michal <mictadlo at gmail.com> wrote:
> Hi Peter,
> Do you have the script which read
>
> https://bitbucket.org/galaxy/galaxy-central/src/8eaf07a46623/test-data/blastp_four_human_vs_rhodopsin.xml
>
>
> and what would be the correct output?
>
> Thank you in advance.
>
> Cheers,
> Michal

Hi Michal,

I'm not quite sure what you're asking, but I'll try. First, the three
data files:

$ wget https://bitbucket.org/galaxy/galaxy-central/src/8eaf07a46623/test-data/blastp_four_human_vs_rhodopsin.xml
$ wget https://bitbucket.org/galaxy/galaxy-central/src/8eaf07a46623/test-data/four_human_proteins.fasta
$ wget https://bitbucket.org/galaxy/galaxy-central/src/8eaf07a46623/rhodopsin_proteins.fasta

The query file has four sequences,

$ grep -c "^>" four_human_proteins.fasta
4

$ grep "^>" four_human_proteins.fasta
>sp|Q9BS26|ERP44_HUMAN Endoplasmic reticulum resident protein 44 OS=Homo sapiens GN=ERP44 PE=1 SV=1
>sp|Q9NSY1|BMP2K_HUMAN BMP-2-inducible protein kinase OS=Homo sapiens GN=BMP2K PE=1 SV=2
>sp|P06213|INSR_HUMAN Insulin receptor OS=Homo sapiens GN=INSR PE=1 SV=4
>sp|P08100|OPSD_HUMAN Rhodopsin OS=Homo sapiens GN=RHO PE=1 SV=1

Based on past experience, I would expect 4 iteration blocks in the
XML, but in this case I have 24:

$ grep "<Iteration>" -c blastp_four_human_vs_rhodopsin.xml
24

Notice we get 6 iterations for each query (4 times 6 is 24):

$ grep "<Iteration_query-ID>" blastp_four_human_vs_rhodopsin.xml
      <Iteration_query-ID>sp|Q9BS26|ERP44_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|Q9BS26|ERP44_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|Q9BS26|ERP44_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|Q9BS26|ERP44_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|Q9BS26|ERP44_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|Q9BS26|ERP44_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|Q9NSY1|BMP2K_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|Q9NSY1|BMP2K_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|Q9NSY1|BMP2K_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|Q9NSY1|BMP2K_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|Q9NSY1|BMP2K_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|Q9NSY1|BMP2K_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P06213|INSR_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P06213|INSR_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P06213|INSR_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P06213|INSR_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P06213|INSR_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P06213|INSR_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P08100|OPSD_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P08100|OPSD_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P08100|OPSD_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P08100|OPSD_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P08100|OPSD_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P08100|OPSD_HUMAN</Iteration_query-ID>

Now, using the two FASTA files directly and re-running blastp, what do I get?

$ ~/Downloads/ncbi-blast-2.2.25+/bin/blastp -query
four_human_proteins.fasta -subject rhodopsin_proteins.fasta -outfmt 5
| grep "<Iteration>" -c
24

Or again with -parse_deflines, which changes how the hit ID/def is presented:

$ ~/Downloads/ncbi-blast-2.2.25+/bin/blastp -query
four_human_proteins.fasta -subject rhodopsin_proteins.fasta -outfmt 5
-parse_deflines | grep "<Iteration>" -c
24

How about older versions?

$ ~/Downloads/ncbi-blast-2.2.24+/bin/blastp -query
four_human_proteins.fasta -subject rhodopsin_proteins.fasta -outfmt 5
BLAST engine error: XML formatting is only supported for a database search

I'll have to make a blast database first...

$ ~/Downloads/ncbi-blast-2.2.24+/bin/makeblastdb -in
rhodopsin_proteins.fasta -dbtype prot

Building a new DB, current time: 05/04/2011 11:22:57
New DB name:   rhodopsin_proteins.fasta
New DB title:  rhodopsin_proteins.fasta
Sequence type: Protein
Keep Linkouts: T
Keep MBits: T
Maximum file size: 1073741824B
Adding sequences from FASTA; added 6 sequences in 0.105655 seconds.

$ ~/Downloads/ncbi-blast-2.2.25+/bin/blastp -query
four_human_proteins.fasta -db rhodopsin_proteins.fasta -outfmt 5 |
grep "<Iteration>" -c
4

Look - just four identifiers as I expect! This also works if the database
is built with the -parse_seqids switch.

The same happens with older versions of BLAST+, one <Iteration>
block per query, so four iteration blocks for this example. I tried all
of 2.2.21+, 2.2.22+, 2.2.23+ and 2.2.24+ (running makeblastdb to
give a fresh database, then blastp).

That seems to demonstrate that bug is specific to the XML output
from FASTA vs FASTA (not FASTA vs DB), which is a new feature
in NCBI BLAST 2.2.25+

I will raise this with the NCBI, and report back.

However, even if the NCBI fix it in the next release, we (Bio*) may
want to update our parsers to cope with this quirk, or at least put a
warning in our BLAST XML parser documentation, as there will be
lots of installations of NCBI BLAST 2.2.25+ in the wild.

Peter

From p.j.a.cock at googlemail.com  Tue May  3 09:24:08 2011
From: p.j.a.cock at googlemail.com (Peter Cock)
Date: Tue, 3 May 2011 10:24:08 +0100
Subject: [Open-bio-l] Interesting BLAST 2.2.25+ XML behaviour
In-Reply-To: <BANLkTi=6_2bFpGhOwxtdjy-DzxUotVWxEg@mail.gmail.com>
References: <BANLkTi=6_2bFpGhOwxtdjy-DzxUotVWxEg@mail.gmail.com>
Message-ID: <BANLkTinH0y4KQ7_AXt7Ly3TgN9fXxErUzA@mail.gmail.com>

Hello all,

I've CC'd the BioPerl, BioRuby, BioJava and Biopython development mailing
lists to make sure you're aware of this, but can we continue any discussion
on the cross-project open-bio-l mailing list please?

I noticed that recent versions of BLAST are not using a single <iteration>
block for each query, which was the historical behaviour and assumed
by the Biopython BLAST XML parser. This may be a bug in BLAST.
See link below for an example.

Has anyone else noticed this, and has it been reported to the NCBI yet?

Thanks,

Peter

(Not for the first time, I wish there was a public bug tracker for BLAST,
or at least a private bug tracker so we could talk about issues with an
NCBI assigned reference number.)

---------- Forwarded message ----------
From: Peter Cock <p.j.a.cock at googlemail.com>
Date: Wed, Apr 20, 2011 at 6:08 PM
Subject: Interesting BLAST 2.2.25+ XML behaviour
To: Biopython-Dev Mailing List <biopython-dev at biopython.org>


Hi all,

Have a look at this XML file from a FASTA vs FASTA search
using blastp from ?BLAST 2.2.25+ (current release), which
is a test file I created for the BLAST+ wrappers in Galaxy:

https://bitbucket.org/galaxy/galaxy-central/src/8eaf07a46623/test-data/blastp_four_human_vs_rhodopsin.xml

I just put it though the Biopython BLAST XML parser, and
was surprised not to get four records back (since as you
might guess from the filename, there were four queries).

It appears this version of BLAST+ is incrementing the
iteration counter for each match... or something like that.

Has anyone else noticed this? I wonder if it is accidental...

Peter



From cjfields at illinois.edu  Tue May  3 13:31:55 2011
From: cjfields at illinois.edu (Chris Fields)
Date: Tue, 3 May 2011 08:31:55 -0500
Subject: [Open-bio-l] [BioRuby] Interesting BLAST 2.2.25+ XML behaviour
In-Reply-To: <BANLkTinH0y4KQ7_AXt7Ly3TgN9fXxErUzA@mail.gmail.com>
References: <BANLkTi=6_2bFpGhOwxtdjy-DzxUotVWxEg@mail.gmail.com>
	<BANLkTinH0y4KQ7_AXt7Ly3TgN9fXxErUzA@mail.gmail.com>
Message-ID: <398303E2-1195-4CC2-8B73-09C6C1117892@illinois.edu>

Haven't tried this using the latest BLAST+ myself, but it doesn't surprise me too much.  Also agree re: some kind of bug tracking with NCBI; I believe they have an internal one, but it would be nice to have a public interface to it.

chris

On May 3, 2011, at 4:24 AM, Peter Cock wrote:

> Hello all,
> 
> I've CC'd the BioPerl, BioRuby, BioJava and Biopython development mailing
> lists to make sure you're aware of this, but can we continue any discussion
> on the cross-project open-bio-l mailing list please?
> 
> I noticed that recent versions of BLAST are not using a single <iteration>
> block for each query, which was the historical behaviour and assumed
> by the Biopython BLAST XML parser. This may be a bug in BLAST.
> See link below for an example.
> 
> Has anyone else noticed this, and has it been reported to the NCBI yet?
> 
> Thanks,
> 
> Peter
> 
> (Not for the first time, I wish there was a public bug tracker for BLAST,
> or at least a private bug tracker so we could talk about issues with an
> NCBI assigned reference number.)
> 
> ---------- Forwarded message ----------
> From: Peter Cock <p.j.a.cock at googlemail.com>
> Date: Wed, Apr 20, 2011 at 6:08 PM
> Subject: Interesting BLAST 2.2.25+ XML behaviour
> To: Biopython-Dev Mailing List <biopython-dev at biopython.org>
> 
> 
> Hi all,
> 
> Have a look at this XML file from a FASTA vs FASTA search
> using blastp from  BLAST 2.2.25+ (current release), which
> is a test file I created for the BLAST+ wrappers in Galaxy:
> 
> https://bitbucket.org/galaxy/galaxy-central/src/8eaf07a46623/test-data/blastp_four_human_vs_rhodopsin.xml
> 
> I just put it though the Biopython BLAST XML parser, and
> was surprised not to get four records back (since as you
> might guess from the filename, there were four queries).
> 
> It appears this version of BLAST+ is incrementing the
> iteration counter for each match... or something like that.
> 
> Has anyone else noticed this? I wonder if it is accidental...
> 
> Peter
> 
> _______________________________________________
> BioRuby Project - http://www.bioruby.org/
> BioRuby mailing list
> BioRuby at lists.open-bio.org
> http://lists.open-bio.org/mailman/listinfo/bioruby




From p.j.a.cock at googlemail.com  Wed May  4 10:36:57 2011
From: p.j.a.cock at googlemail.com (Peter Cock)
Date: Wed, 4 May 2011 11:36:57 +0100
Subject: [Open-bio-l] [BioRuby] Interesting BLAST 2.2.25+ XML behaviour
In-Reply-To: <4DC12371.3040204@gmail.com>
References: <BANLkTi=6_2bFpGhOwxtdjy-DzxUotVWxEg@mail.gmail.com>
	<BANLkTinH0y4KQ7_AXt7Ly3TgN9fXxErUzA@mail.gmail.com>
	<398303E2-1195-4CC2-8B73-09C6C1117892@illinois.edu>
	<4DC12371.3040204@gmail.com>
Message-ID: <BANLkTinV4uha74Y9jC_f=XLK5LufAn1xHw@mail.gmail.com>

On Wed, May 4, 2011 at 10:59 AM, Michal <mictadlo at gmail.com> wrote:
> Hi Peter,
> Do you have the script which read
>
> https://bitbucket.org/galaxy/galaxy-central/src/8eaf07a46623/test-data/blastp_four_human_vs_rhodopsin.xml
>
>
> and what would be the correct output?
>
> Thank you in advance.
>
> Cheers,
> Michal

Hi Michal,

I'm not quite sure what you're asking, but I'll try. First, the three
data files:

$ wget https://bitbucket.org/galaxy/galaxy-central/src/8eaf07a46623/test-data/blastp_four_human_vs_rhodopsin.xml
$ wget https://bitbucket.org/galaxy/galaxy-central/src/8eaf07a46623/test-data/four_human_proteins.fasta
$ wget https://bitbucket.org/galaxy/galaxy-central/src/8eaf07a46623/rhodopsin_proteins.fasta

The query file has four sequences,

$ grep -c "^>" four_human_proteins.fasta
4

$ grep "^>" four_human_proteins.fasta
>sp|Q9BS26|ERP44_HUMAN Endoplasmic reticulum resident protein 44 OS=Homo sapiens GN=ERP44 PE=1 SV=1
>sp|Q9NSY1|BMP2K_HUMAN BMP-2-inducible protein kinase OS=Homo sapiens GN=BMP2K PE=1 SV=2
>sp|P06213|INSR_HUMAN Insulin receptor OS=Homo sapiens GN=INSR PE=1 SV=4
>sp|P08100|OPSD_HUMAN Rhodopsin OS=Homo sapiens GN=RHO PE=1 SV=1

Based on past experience, I would expect 4 iteration blocks in the
XML, but in this case I have 24:

$ grep "<Iteration>" -c blastp_four_human_vs_rhodopsin.xml
24

Notice we get 6 iterations for each query (4 times 6 is 24):

$ grep "<Iteration_query-ID>" blastp_four_human_vs_rhodopsin.xml
      <Iteration_query-ID>sp|Q9BS26|ERP44_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|Q9BS26|ERP44_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|Q9BS26|ERP44_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|Q9BS26|ERP44_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|Q9BS26|ERP44_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|Q9BS26|ERP44_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|Q9NSY1|BMP2K_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|Q9NSY1|BMP2K_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|Q9NSY1|BMP2K_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|Q9NSY1|BMP2K_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|Q9NSY1|BMP2K_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|Q9NSY1|BMP2K_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P06213|INSR_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P06213|INSR_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P06213|INSR_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P06213|INSR_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P06213|INSR_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P06213|INSR_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P08100|OPSD_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P08100|OPSD_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P08100|OPSD_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P08100|OPSD_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P08100|OPSD_HUMAN</Iteration_query-ID>
      <Iteration_query-ID>sp|P08100|OPSD_HUMAN</Iteration_query-ID>

Now, using the two FASTA files directly and re-running blastp, what do I get?

$ ~/Downloads/ncbi-blast-2.2.25+/bin/blastp -query
four_human_proteins.fasta -subject rhodopsin_proteins.fasta -outfmt 5
| grep "<Iteration>" -c
24

Or again with -parse_deflines, which changes how the hit ID/def is presented:

$ ~/Downloads/ncbi-blast-2.2.25+/bin/blastp -query
four_human_proteins.fasta -subject rhodopsin_proteins.fasta -outfmt 5
-parse_deflines | grep "<Iteration>" -c
24

How about older versions?

$ ~/Downloads/ncbi-blast-2.2.24+/bin/blastp -query
four_human_proteins.fasta -subject rhodopsin_proteins.fasta -outfmt 5
BLAST engine error: XML formatting is only supported for a database search

I'll have to make a blast database first...

$ ~/Downloads/ncbi-blast-2.2.24+/bin/makeblastdb -in
rhodopsin_proteins.fasta -dbtype prot

Building a new DB, current time: 05/04/2011 11:22:57
New DB name:   rhodopsin_proteins.fasta
New DB title:  rhodopsin_proteins.fasta
Sequence type: Protein
Keep Linkouts: T
Keep MBits: T
Maximum file size: 1073741824B
Adding sequences from FASTA; added 6 sequences in 0.105655 seconds.

$ ~/Downloads/ncbi-blast-2.2.25+/bin/blastp -query
four_human_proteins.fasta -db rhodopsin_proteins.fasta -outfmt 5 |
grep "<Iteration>" -c
4

Look - just four identifiers as I expect! This also works if the database
is built with the -parse_seqids switch.

The same happens with older versions of BLAST+, one <Iteration>
block per query, so four iteration blocks for this example. I tried all
of 2.2.21+, 2.2.22+, 2.2.23+ and 2.2.24+ (running makeblastdb to
give a fresh database, then blastp).

That seems to demonstrate that bug is specific to the XML output
from FASTA vs FASTA (not FASTA vs DB), which is a new feature
in NCBI BLAST 2.2.25+

I will raise this with the NCBI, and report back.

However, even if the NCBI fix it in the next release, we (Bio*) may
want to update our parsers to cope with this quirk, or at least put a
warning in our BLAST XML parser documentation, as there will be
lots of installations of NCBI BLAST 2.2.25+ in the wild.

Peter