What kinase phosphorylates WNK2 at site s49 (and s45) ? And how to deal with chatGPT with such questions ?

Question

[Alexander]

Question: What kinase(s) phosphorylates WNK2 at site s49 (and s45) ?

1) Googling does not help me. 2) Nice site:https://www.phosphosite.org/proteinAction.action?id=2242&showAllSites=true - I cannot find any information what kinases phosphorilates sites - I see only sites, but not what kinases creates the phosphorilation. 3) Uniprot provides information on "s45", but not on s49 - look on section "ptm processing" https://www.uniprot.org/uniprotkb/Q9Y3S1/entry#ptm_processing - go to "Modified residue 45" - there is a paper: https://pubmed.ncbi.nlm.nih.gov/21733846/ and FIGURE 6 (table) in it with "Predicted Kinases" DMPK, PAK. It is not clear whether these predictions are reliable and are there any other opinions. 4) some other data sources including omnipath does not contain information on WNK2 at all.

==========

So if you already played with chatGPT, you may enjoy its answers, it is natural to appeal to it for such question. It is really great source of knowledge. However there are some subtle points.

1) You cannot completely trust chatGPT - it is well-known that it can describe very smoothly and convincingly quite a nonsense. Second what seems to me - it can mix some details - I am afraid for example it can give information for say WNK4, instead of WNK2 , but mentioning WNK2.

2) It may not directly answer the question. "what kinase phosphorilates protein WNK2 at positions S45, S49" - does not provide a precise answer. But for that there is a trick: A) ask a more general question - "what kinases phosphorilates protein WNK2 at any position ?" B) after that go to specific question: "please give me positions where these kinases phosphorilate the WNK2" In a sense it is like with human - first explain the context, than go to details. But it strange to see such from machine.

Full dialogue is saved here: https://www.kaggle.com/code/alexandervc/mb-21a-search-wnk-in-kinase-target-data?scriptVersionId=116937086&cellId=3

And you get the answer from ChatGPT:

The specific positions on the WNK2 protein where these kinases phosphorylate it may vary depending on the cellular context and the conditions of the experiment.
However, some of the known phosphorylation sites on WNK2 are:

Protein Kinase A (PKA) phosphorylates WNK2 at S45 and S49

Protein Kinase C (PKC) phosphorylates WNK2 at S45 and S49

Mitogen-activated protein kinase (MAPK) phosphorylates WNK2 at T365

Protein Kinase G (PKG) phosphorylates WNK2 at S49

Protein Kinase D (PKD) phosphorylates WNK2 at T365

Cyclin-dependent kinase (CDK) phosphorylates WNK2 at T365

Casein kinase 2 (CK2) phosphorylates WNK2 at S49

GSK-3 (Glycogen synthase kinase 3) phosphorylates WNK2 at S49

It is important to note that these are examples of known phosphorylation sites and that the specific phosphorylation site may depend on the specific condition of the experiment.

===========

Question: Is that information reliable ? What are your general advices to deal with chatGPT - to verify its answers ? to force to give your answers when it does not not answer directly at first ?

[Joe]

I think the answer is pretty simple - do what we've always done and lean on the scientific literature.

DO NOT use ChatGPT answers whole cloth without extensively fact-checking them, in particular, one thing your post doesn't mention is that ChatGPT will completely make up references.

Answer 1

[Michael]

My attempt to teach ChatGPT to translate a sequence, any more questions?

I need to translate a nucleotide sequence into amino acid sequence, can you do that?

Yes, I can help you translate a nucleotide sequence into an amino acid sequence. The process is called "protein translation" and it occurs in cells through the process of transcription and translation.

The genetic code is read in groups of three nucleotides, called codons. Each codon corresponds to a specific amino acid or to a stop signal. There is a one-to-one correspondence between the codons and the amino acids, so the nucleotide sequence of a gene can be translated into the amino acid sequence of a protein.

Can you provide the nucleotide sequence you would like translated?

Sure: AUGUUUUAG

The amino acid sequence that corresponds to the nucleotide sequence "AUGUUUUAG" is:

M (Methionine) - F (Phenylalanine) - F (Phenylalanine) - Stop

AUG codes for methionine, UUU codes for phenylalanine and AG codes for stop. Please note that this is a small sequence, in real world scenario the sequence could be longer and also there could be more than one reading frame.

How many nucleotides does a codon contain?

A codon is a sequence of three nucleotides that codes for a specific amino acid or a stop signal during protein translation. This means that each codon contains three nucleotides.

The genetic code is read in groups of three nucleotides, called codons. Each codon corresponds to a specific amino acid or to a stop signal. There are 64 possible codons, 61 of which code for amino acids and 3 that signal the end of a translation.

Ok, now in light of that information, how many codons did my input sequence contain?

The input sequence "AUGUUUUAG" contains 3 codons.

AUG codes for Methionine ,UUU codes for Phenylalanine and AG codes for stop.

Each codon is composed of 3 nucleotides and this input sequence contains 3 codons, each one of them is composed of 3 nucleotides.

You say AG codes for stop? Could you please double-check that information?

I apologize for the confusion. AG does not code for stop, it codes for the amino acid arginine. The codons that signal the end of a translation are UAA, UAG, and UGA, these are called stop codons or termination codons.

In the input sequence "AUGUUUUAG" the last codon AG codes for arginine and not stop. Please let me know if you have any further questions.

Ok, you just said AG codes for an amino acid, but a codon is of length three. Do you notice your mistake?

I apologize, my previous response was incorrect. You are correct that a codon is a sequence of three nucleotides and that the input sequence "AUGUUUUAG" is not a multiple of three, therefore it cannot be translated into an amino acid sequence.

A codon is a sequence of three nucleotides that codes for a specific amino acid or a stop signal during protein translation. These codons are read in a sequence starting from the first base of a mRNA called the AUG start codon and ending on a stop codon (UAA, UAG, UGA). An input sequence must be multiple of 3 nucleotides long in order to be translated.

I apologize for any confusion my previous answers may have caused. Please let me know if you have any further questions.

what is the length of the sequence AUGUUUUAG ?

The input sequence "AUGUUUUAG" is 8 nucleotides long. Each nucleotide is a building block of DNA and RNA, it could be adenine (A), thymine (T), cytosine (C) or guanine (G) in the case of DNA or adenine (A), uracil (U), cytosine (C) or guanine (G) in the case of RNA. As this sequence is not a multiple of 3, it cannot be translated into an amino acid sequence.

I am sorry but that is getting worse and worse. The length of my example is 9 nucleotides.

I apologize for my mistake. You are correct that the input sequence "AUGUUUUAG" is 9 nucleotides long. I apologize for any confusion my previous answers may have caused. It's a common mistake to assume that the input sequence is the coding region or an ORF (Open Reading Frame) which is the region of a DNA or RNA molecule that can be translated to a protein. The coding region of a gene is a subregion of the gene that contains information for the synthesis of a functional product such as a protein, and it is usually multiple of three. Therefore, it is important to note that the input sequence might not be a coding region or an ORF and it cannot be translated into an amino acid sequence. Please let me know if you have any further questions.

Ok, in the light of what we have just learned, how many codons does the sequence comprise of?

The input sequence "AUGUUUUAG" is 9 nucleotides long, and as we know that a codon is a sequence of three nucleotides, so it will comprise of 3 codons. It would be AUG - UUU - UUA Please note that this sequence is not a coding sequence and it cannot be translated into an amino acid sequence. Let me know if you have any other question.

I am sorry, this is hopeless...

I apologize for any confusion my previous responses may have caused. As a language model AI, I can make mistakes and provide incorrect information, but I strive to provide accurate and helpful responses. If you have any other question or if there is any other way I can assist you, please let me know.

[harold.smith.tarheel]

Permission to share?

[Michael]

Sure, or what do you mean? Note that this is not meant to diminish the achievement of ChatGPT in natural language generation and semantic processing in any way, just to demonstrate some limitations.

Also, it is able to follow instructions, so it should be possible to "program" it to do the translation right, like a task of code golf.

[Michael]

Here is a counter-example, where we actually make it program itself:

Human: Given a nucleotide sequence, please explain how to translate it into amino acid in bullet points

ChatGPT: Sure, here are the steps to translate a nucleotide sequence into an amino acid sequence:

1. Divide the nucleotide sequence into groups of three nucleotides, called codons.
2. Look up the amino acid that corresponds to each codon using the genetic code, which is the relationship between codons and amino acids.
3. Assemble the amino acids in the order they were read from the nucleotide sequence to form the amino acid sequence. Stop when a stop codon is encountered.

It's worth noting that the process of translation is carried out by a cell's machinery, including molecules called ribosomes and transfer RNAs (tRNAs), and the steps above are a simplified version of the process.

H: Here are the steps to translate a nucleotide sequence into an amino acid sequence: Divide the nucleotide sequence into groups of three nucleotides, called codons. Look up the amino acid that corresponds to each codon using the genetic code, which is the relationship between codons and amino acids. Assemble the amino acids in the order they were read from the nucleotide sequence to form the amino acid sequence. Stop when a stop codon is encountered.

C: Yes, that's correct! Divide the nucleotide sequence [...]

It's worth noting that the process of translation is a complex process that is carried out by a cell's machinery, including molecules called ribosomes and transfer RNAs (tRNAs) and the steps above are a simplified version of the process.

H: Please now apply the instructions above to the following sequence: ATGTTTTAG

C: Sure, here's the process of translating the nucleotide sequence "ATGTTTTAG" into an amino acid sequence:

Divide the nucleotide sequence into groups of three nucleotides, called codons. In this case, the codons would be "ATG", "TTT", "TAG" Look up the amino acid that corresponds to each codon using the genetic code. "ATG" corresponds to methionine, "TTT" corresponds to phenylalanine, "TAG" is a stop codon, indicating the end of translation.

Assemble the amino acids in the order they were read from the nucleotide sequence to form the amino acid sequence. The amino acid sequence is methionine-phenylalanine

Stop when a stop codon is encountered.

Keep in mind that this is a simplified version of the translation process, which is carried out by a cell's machinery, including molecules called ribosomes and transfer RNAs (tRNAs).

[harold.smith.tarheel]

Thanks. It's useful to have a concrete example such as this one to demonstrate some of the limitations of ChatGPT (e.g., the combination of accurate and inaccurate information, the confident statement of inaccuracies, etc).

Answer 2

[andreasstuermer]

ChatGPT once told me that the information I was looking for was written in a paper (the title was credible, the authors names were credible, the years was credible).

The paper did not exist. Not on google, not on google scholar.

I assume it invented that a paper exists. In reality, It dd not seem like someone ever did the experiment to answer the question I was asking.

[Alexander]

Thank you for sharing striking example ! It seems to me recently developers forbidded chatGPT to provide references, at least I cannot enforce it to give me any.

PS

Just some more experiment shows that probably providing references is not shut down completely, at least when the answer of chatGPT was like "has been shown... in some STUDIES", asking the next question: "what are these studies?" provide some (not-detailed) reference - like title, year journal. (No authors). But reference indeed appears to be completely fake.