5
$\begingroup$

There are certain proteins that contain metal components, known as metalloproteins. Commonly, the metal is at the active site which needs the most prediction precision. Typically, there is only one (or a few) metals in a protein, which contains far more other atoms. So, the structural data that we could be used to train AlphaFold will contain far less information about the metal elements. Not to mention most proteins don't have metals at all (it is estimated that only 1/2-1/4 of all proteins contain metals [1]).

Given that maybe there is not enough structural data about protein local structure around metal atoms (e.g. Fe, Zn, Mg, etc.), then AlphaFold cannot predict local structure around metals well. Is that right?

I also think that the more complex electron shell of metal also makes the data less useful, since its bounding pattern is more flexible than carbon, etc.

$\endgroup$
2
  • 1
    $\begingroup$ I've edited your post to add more details based on your comments. Can you explain this sentence "Commonly, the metal is at the active site which needs the most prediction precision"? Why do you say that it needs "the most prediction precision"? Maybe you could look at the exact proteins used to train the AlphaFold in the CASP competition. Maybe there's some info about the type of proteins used to train AlphaFold. $\endgroup$
    – nbro
    Dec 6 '20 at 14:46
  • 1
    $\begingroup$ @nbro The prediction precision for the active-site will be much more important than for other parts of the protein, because the active site is where the most "action" takes place. $\endgroup$ Dec 6 '20 at 21:50
4
$\begingroup$

Let me address first some of the things you wrote in your question:

There are certain proteins that contain metal components, known as metalloproteins.

Natural proteins do not ever contain metal components as far as we know. Natural proteins are composed of natural amino acids which only contain H,C,O,N,S. Selenocysteine contains Se (also a non-metal!) but it's a proteinogenic amino acid, which means it's a precursor to proteins but doesn't typically show up in the protein itself. From the Wikipedia page that you gave us in your question: "Metalloprotein is a generic term for a protein that contains a metal ion cofactor" and "A cofactor is a non-protein chemical compound or metallic ion".

But your question still deserves an answer, because even though it would incorrect to call a co-factor "part" of the protein, they can still affect the folding and the overall shape. But let's first address one last part of your question:

Given that maybe there is not enough structural data about protein local structure around metal atoms (e.g. Fe/Zn/Mg), then AlphaFold cannot predict local structure around metals well. Is that right?

The first sentence of the Wikipedia article that you linked in your question, says "For instance, at least 1000 human proteins (out of ~20,000) contain zinc-binding protein domains [3] although there may be up to 3000 human zinc metalloproteins [4]." Therefore, while metalloproteins might not be the majority of all proteins, there's a decent enough number of them that are of relevance to the human body, and therefore constructing training databases that contain enough metalloproteins (or even 100% metalloproteins, if desired) is not difficult.

I mentioned a bit elsewhere that AlphaFold was used to predict protein structures in the CASP competition, for which you can see for yourself that many/most of the proteins for which contestants (such as DeepMind) need to predict the structure, come from studies of proteins of relevance to humans because the CASP structures typically come from X-ray crystallography studies, which are typically done on proteins of relevance to humans.

You can also see for yourself not only the "target list" that I showed above, but also the results of the competition which will show how well AlphaFold performed in CASP13 (2018) and CASP14 (2020) for metalloproteins.

Finally:

I also think that the more complex electron shell of metal also makes the data less useful, since its bounding pattern is more flexible than carbon, etc.

It is true that metals are typically harder to model than C,H,O,N,S, and even Se, if doing ab initio calculations on the metals or metail-containing complexes. However, the purpose of machine learning in protein folding studies is to skip ab initio, statistical-dynamical and/or molecular-dynamical calculations of the relevant structures and simply use training data to predict the protein structures. That being said, there needs to be enough training data available (as you correctly pointed out) to learn what happens near the metal co-factors: The answer to this is that there are indeed enough metalloproteins to sufficiently populate a training set, but they won't contain enough of the specific metals involved in every metalloprotein. For example, lots of data will be available for proteins containing Fe since Fe is in hemoglobin (for example) which is essential to the functioning of red blood cells to absorb oxygen, but the protein vanabins contains vanadium which is much rarer and therefore training data involving it will be much less available. You're also correct that metal elements can form more bonds than typical elements found in organic compounds.

So it depends on the metal in the relevant co-factor. Fe-based co-factors will have quite a lot of training data available, as well as Mg-based ones, Zn-based ones, and a lot of other ones which contain the "more common" metals. For proteins like vanabins which contains vanadium, you are quite correct that training data will be limited, but also keep in mind that vanabins is a very rare protein found in sea squirts and we already know more about its structure (through X-ray crystallography, which means we don't need machine learning for it) than we even know about what it even does. The chances of other vanadium-containing co-factors in metalloproteins being very significant is too low to justify working on protein folding algorithms specifically for them.

$\endgroup$
7
  • $\begingroup$ The part "There are certain proteins that contain metal components, known as metalloproteins." was written by me to provide some introduction, so, please, do not address that in this answer, but please propose an edit to the original post to fix that description (if you really think it's wrong). I wrote "parts" to provide a generic/general description, given that the OP had not provided any description initially. $\endgroup$
    – nbro
    Dec 6 '20 at 22:59
  • $\begingroup$ Well you shouldn't be injecting scientifically incorrect edits into other people's posts if you don't know enough about the topic. I just spent a couple hours on the answer and got myself late for something, so I was about to leave and will have to make those edits later. $\endgroup$ Dec 6 '20 at 23:01
  • $\begingroup$ Where did I inject scientifically wrong info? The first sentence of the first section of that article says "It is estimated that approximately half of all proteins contain a metal.". I wrote "components" and "parts" and I don't know why you think that's wrong as a generic (of course, not very precise) description (but this site is not about biological concepts, so you should expect some imprecision). See also en.wikipedia.org/wiki/Rubredoxin, for example. I also spent a lot of time trying to clarify the original question, which was not answerable at all in the beginning. $\endgroup$
    – nbro
    Dec 6 '20 at 23:05
  • $\begingroup$ So, not sure where your conclusion "Natural proteins do not ever contain metal components as far as we know" comes from. For instance, here we have "Metalloproteins are commonplace in biological systems.". It's true that I am not expert in this specific topic, but I studied biology in high school a few years ago, so my knowledge of biological concepts is not completely zero. I have an idea of what an atom, ion, protein, amino acid, etc., are. $\endgroup$
    – nbro
    Dec 6 '20 at 23:10
  • 1
    $\begingroup$ IMO cofactor has two types, first is a protein or small molecule that use normal weak interaction to bind to the main protein. The second is metal like, which use metal-sepecific coordination bound to bind to the main protein. Although all called cofactor, but they are not very same. IMO the metal cofactor is better to be considered as a must-have structural component of the protein, not a "factor" that modify a protein's properties when bound $\endgroup$
    – jw_
    Dec 6 '20 at 23:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.