{ "id": 429218, "title": "Inconsistent Protein Sequences for the Same Protein", "url": "/competitions/cafa-5-protein-function-prediction/discussion/429218", "author_name": "", "post_date": "2023-08-04T13:53:02.606487600Z", "votes": 5, "comment_count": 5, "views": 0, "content": "

Hi everyone,

\n

I currently explore the PDB files downloaded from https://alphafold.ebi.ac.uk and observe there exists inconsistency of protein sequence with the corresponding protein in train_sequences.fasta. Following shows one of the example:

\n

For protein Q9RYE6, there are 231 amino acids in the sequence from train_sequences.fasta,

\n
\n
\n

As for the corresponding one from https://alphafold.ebi.ac.uk, there are 291 amino acids in the sequence,

\n
\n
\n

The code snippet I use to parse the PDB file is as follows:

\n
 Bio  SeqIO\n\nseq_records = SeqIO.parse(<path to PDB file>, )\nseq = [(record.seq)  record  seq_records][]\n
\n

Two sequences look totally different here. Hence, I wonder what's the reason causing this issue. Thanks for the help.

", "messages": [ { "id": "2373907", "postDate": "08/04/2023 13:53:02", "content": "

Hi everyone,

\n

I currently explore the PDB files downloaded from https://alphafold.ebi.ac.uk and observe there exists inconsistency of protein sequence with the corresponding protein in train_sequences.fasta. Following shows one of the example:

\n

For protein Q9RYE6, there are 231 amino acids in the sequence from train_sequences.fasta,

\n
\n
\n

As for the corresponding one from https://alphafold.ebi.ac.uk, there are 291 amino acids in the sequence,

\n
\n
\n

The code snippet I use to parse the PDB file is as follows:

\n
 Bio  SeqIO\n\nseq_records = SeqIO.parse(<path to PDB file>, )\nseq = [(record.seq)  record  seq_records][]\n
\n

Two sequences look totally different here. Hence, I wonder what's the reason causing this issue. Thanks for the help.

", "rawMarkdown": "Hi everyone,\n\nI currently explore the PDB files downloaded from `https://alphafold.ebi.ac.uk` and observe there exists inconsistency of protein sequence with the corresponding protein in `train_sequences.fasta`. Following shows one of the example:\n\nFor protein `Q9RYE6`, there are 231 amino acids in the sequence from `train_sequences.fasta`,\n```\nMKNAPLTLNFGSVRLPVSADGLLHAPTAQQQLGLTQSWEAALVEHGLPETYRDFGAGPEAAVSVPDFVALAFALDTPEARRWQKRARELLARAMQGDVRVAAQIAERNPEPDARRWLAARLESTGARRELLATVARHGGEGRVYGQLGSISNRTVLGKDSASVRQERGVKATRDGLTSAELLRLAYIDTVTARAIQESEARGNAAILTLHEQVARSERQSWERAGQVQRVG\n```\n\nAs for the corresponding one from `https://alphafold.ebi.ac.uk`, there are 291 amino acids in the sequence,\n```\nMGTGDPSPLASQGPLPLVEGQKMKKPPPVRRGMNEAMEDRGSFFMALVSAYALHLARPFPALALAAGYLFVEGEDGCVAPRFAFLNGAGGHGVDVGQPQQFGARQAVAGGLDAAFLTYRRAVLAKDGAVADAAELTVDPALATVPGDGGEQFAPGASAFQSGSEPAPGVRLGVALGNLGSHAHVALHGAGEQLPRPLLPAPGFGGVEGEGKRHEIGHAHGGLGSGAEVAVRLGQAVFDQRRFPALGEPELLLGGGGVKQTVRADRQPHGAEIEGQRSVLHTDILAQIMFLT\n```\n\nThe code snippet I use to parse the PDB file is as follows:\n```python\nfrom Bio import SeqIO\n\nseq_records = SeqIO.parse(, \"pdb-atom\")\nseq = [str(record.seq) for record in seq_records][0]\n```\n\nTwo sequences look totally different here. Hence, I wonder what's the reason causing this issue. Thanks for the help.", "votes": null }, { "id": "2373967", "postDate": "08/04/2023 14:43:30", "content": "

Good question, I'm afraid I don't have an answer. Sometimes it's a question of a local alignment with some N or C terminal residues missing. But here the sequences are completely different. I already filed a query with DeepMind, and I'll ask uniprot.

", "rawMarkdown": "Good question, I'm afraid I don't have an answer. Sometimes it's a question of a local alignment with some N or C terminal residues missing. But here the sequences are completely different. I already filed a query with DeepMind, and I'll ask uniprot.", "votes": null }, { "id": "2373979", "postDate": "08/04/2023 14:56:48", "content": "

Hi @iddux1,

\n

Yeh, I found some proteins have only minor inconsistency, for example, missing the first 6 amino acids with the remaining exactly matched.
\nThanks for your quick reply and effort!

", "rawMarkdown": "Hi @iddux1,\n\nYeh, I found some proteins have only minor inconsistency, for example, missing the first 6 amino acids with the remaining exactly matched. \nThanks for your quick reply and effort!", "votes": null }, { "id": "2374175", "postDate": "08/04/2023 18:34:20", "content": "

That sequence was revised in Dec 2022 - after the original AFDB model was created. If you look the accession code up on the UniProt web site and click on the \"history\" tab you can see what revisions to the sequences have happened. More than you might imagine!

", "rawMarkdown": "That sequence was revised in Dec 2022 - after the original AFDB model was created. If you look the accession code up on the UniProt web site and click on the \"history\" tab you can see what revisions to the sequences have happened. More than you might imagine!", "votes": null }, { "id": "2374266", "postDate": "08/04/2023 19:52:49", "content": "

Revisions happen a lot, but usually it's about in-frame changes. This seems to be a completely different sequence. But I think I found the reason:

\n

Sequence caution:
\n\"The sequence AAF09602.1 differs from that shown. Reason: Miscellaneous discrepancy Was originally predicted on the opposite strand.\"

\n

So the old version of the uniprot sequence was predicted to be coded on the opposite strand, giving a different sequence. Seems like the AlphaFoldDB still uses the old (wrong) sequence.

\n

In the History tab you can see the change between version 75 and 76 of this sequence.

\n

https://www.uniprot.org/uniprotkb/Q9RYE6/history

", "rawMarkdown": "Revisions happen a lot, but usually it's about in-frame changes. This seems to be a completely different sequence. But I think I found the reason:\n\n**Sequence caution:**\n\"The sequence AAF09602.1 differs from that shown. Reason: Miscellaneous discrepancy Was originally predicted on the opposite strand.\"\n\nSo the old version of the uniprot sequence was predicted to be coded on the opposite strand, giving a different sequence. Seems like the AlphaFoldDB still uses the old (wrong) sequence. \n\nIn the History tab you can see the change between version 75 and 76 of this sequence.\n\nhttps://www.uniprot.org/uniprotkb/Q9RYE6/history", "votes": null }, { "id": "2375324", "postDate": "08/05/2023 14:49:42", "content": "

Hi @biodav @iddux1,

\n

Thanks very much for the clarification.

", "rawMarkdown": "Hi @biodav @iddux1,\n\nThanks very much for the clarification.", "votes": null } ], "comments": [ { "id": 2373967, "author_name": "iddux1", "author_url": "", "post_date": "08/04/2023 14:43:30", "content": "

Good question, I'm afraid I don't have an answer. Sometimes it's a question of a local alignment with some N or C terminal residues missing. But here the sequences are completely different. I already filed a query with DeepMind, and I'll ask uniprot.

", "votes": null, "replies": [ { "id": 2373979, "author_name": "abaojiang", "author_url": "", "post_date": "08/04/2023 14:56:48", "content": "

Hi @iddux1,

\n

Yeh, I found some proteins have only minor inconsistency, for example, missing the first 6 amino acids with the remaining exactly matched.
\nThanks for your quick reply and effort!

", "votes": null, "replies": [] } ] }, { "id": 2374175, "author_name": "biodav", "author_url": "", "post_date": "08/04/2023 18:34:20", "content": "

That sequence was revised in Dec 2022 - after the original AFDB model was created. If you look the accession code up on the UniProt web site and click on the \"history\" tab you can see what revisions to the sequences have happened. More than you might imagine!

", "votes": null, "replies": [ { "id": 2374266, "author_name": "iddux1", "author_url": "", "post_date": "08/04/2023 19:52:49", "content": "

Revisions happen a lot, but usually it's about in-frame changes. This seems to be a completely different sequence. But I think I found the reason:

\n

Sequence caution:
\n\"The sequence AAF09602.1 differs from that shown. Reason: Miscellaneous discrepancy Was originally predicted on the opposite strand.\"

\n

So the old version of the uniprot sequence was predicted to be coded on the opposite strand, giving a different sequence. Seems like the AlphaFoldDB still uses the old (wrong) sequence.

\n

In the History tab you can see the change between version 75 and 76 of this sequence.

\n

https://www.uniprot.org/uniprotkb/Q9RYE6/history

", "votes": null, "replies": [ { "id": 2375324, "author_name": "abaojiang", "author_url": "", "post_date": "08/05/2023 14:49:42", "content": "

Hi @biodav @iddux1,

\n

Thanks very much for the clarification.

", "votes": null, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "2373907": "Hi everyone,\n\nI currently explore the PDB files downloaded from `https://alphafold.ebi.ac.uk` and observe there exists inconsistency of protein sequence with the corresponding protein in `train_sequences.fasta`. Following shows one of the example:\n\nFor protein `Q9RYE6`, there are 231 amino acids in the sequence from `train_sequences.fasta`,\n```\nMKNAPLTLNFGSVRLPVSADGLLHAPTAQQQLGLTQSWEAALVEHGLPETYRDFGAGPEAAVSVPDFVALAFALDTPEARRWQKRARELLARAMQGDVRVAAQIAERNPEPDARRWLAARLESTGARRELLATVARHGGEGRVYGQLGSISNRTVLGKDSASVRQERGVKATRDGLTSAELLRLAYIDTVTARAIQESEARGNAAILTLHEQVARSERQSWERAGQVQRVG\n```\n\nAs for the corresponding one from `https://alphafold.ebi.ac.uk`, there are 291 amino acids in the sequence,\n```\nMGTGDPSPLASQGPLPLVEGQKMKKPPPVRRGMNEAMEDRGSFFMALVSAYALHLARPFPALALAAGYLFVEGEDGCVAPRFAFLNGAGGHGVDVGQPQQFGARQAVAGGLDAAFLTYRRAVLAKDGAVADAAELTVDPALATVPGDGGEQFAPGASAFQSGSEPAPGVRLGVALGNLGSHAHVALHGAGEQLPRPLLPAPGFGGVEGEGKRHEIGHAHGGLGSGAEVAVRLGQAVFDQRRFPALGEPELLLGGGGVKQTVRADRQPHGAEIEGQRSVLHTDILAQIMFLT\n```\n\nThe code snippet I use to parse the PDB file is as follows:\n```python\nfrom Bio import SeqIO\n\nseq_records = SeqIO.parse(, \"pdb-atom\")\nseq = [str(record.seq) for record in seq_records][0]\n```\n\nTwo sequences look totally different here. Hence, I wonder what's the reason causing this issue. Thanks for the help.", "2373967": "Good question, I'm afraid I don't have an answer. Sometimes it's a question of a local alignment with some N or C terminal residues missing. But here the sequences are completely different. I already filed a query with DeepMind, and I'll ask uniprot.", "2373979": "Hi @iddux1,\n\nYeh, I found some proteins have only minor inconsistency, for example, missing the first 6 amino acids with the remaining exactly matched. \nThanks for your quick reply and effort!", "2374175": "That sequence was revised in Dec 2022 - after the original AFDB model was created. If you look the accession code up on the UniProt web site and click on the \"history\" tab you can see what revisions to the sequences have happened. More than you might imagine!", "2374266": "Revisions happen a lot, but usually it's about in-frame changes. This seems to be a completely different sequence. But I think I found the reason:\n\n**Sequence caution:**\n\"The sequence AAF09602.1 differs from that shown. Reason: Miscellaneous discrepancy Was originally predicted on the opposite strand.\"\n\nSo the old version of the uniprot sequence was predicted to be coded on the opposite strand, giving a different sequence. Seems like the AlphaFoldDB still uses the old (wrong) sequence. \n\nIn the History tab you can see the change between version 75 and 76 of this sequence.\n\nhttps://www.uniprot.org/uniprotkb/Q9RYE6/history", "2375324": "Hi @biodav @iddux1,\n\nThanks very much for the clarification." }, "source": "meta" }