Code of Conduct
Results 1 to 24 of 24
  1. #1

    Question Aldehydes, So-Called (SMILES)

    Hi guys,

    This is not about the misnamed ACs, but about how/if you can distinguish whether or not an AC is indeed an aldehyde or a ketone from its SMILES string beyond the C=O for the carbonyl group. Has anyone done it algorithmically, here ? Is it possible without having to parse the whole string for the valence tree with something like a regex, instead ?

    Also, does anybody know if the vibrational theory of smell allows for the carbonyl functional group to have a shared characteristic (sometimes dominant?) odor and if it occurs in ACs (with examples, if possible) ? Or if only aldehydes and ketones have shared organoleptics, or only aldhedydes ? Is it verified extensively in practice ? From limited experience, I know that to me, it is the case for most (aptly named) aldehydes (ozonic on top, ashy in the drydown), but I'm not so sure about ketones and even less so with the carbonyl group itself.
    Last edited by contrebande; 15th October 2020 at 01:17 AM. Reason: spelling, readability

  2. #2

    Default Re: Aldehydes, So-Called (SMILES)

    I should know SMILES but don't.

    If the structure has C(=O)H or if you can't see structure the name has -al or formyl, it's an aldehyde. Otherwise not unless there's something I'm not thinking of.

    Aldehydes do not necessarily share organoleptics. Eg C10 aldehyde is not like hexyl cinnamic aldehyde and neither is like Cortex Aldehyde. Benzaldehyde is still different from all these.

  3. #3

    Default Re: Aldehydes, So-Called (SMILES)

    From what I understand, to the vibrational theory of smell, odors receptors act (potentially not exclusively) as spectrometers for a specific bandwidth. Different receptors would be sensitive to different bands of wavelengths, with one AC potentially exciting many receptors to various degrees. We are therefore in a continuous/stochastic realm instead of a binary fit/no fit one as in the shape theory of smell. So "shared organoleptics" might, like I said, not always be dominant or even perceptible. The ozonic organoleptic pattern is indeed less perceptible with cinnamaldehydes, but the ashy drydown is there (to me). And cinnamaldehydes definitely do share "honey" patterns with benzaldehyde and heliothropin and vanillic aldehydes (again, to me). So it might just be that, the continuity and high dimensionality of the euclidian "smellspace" (color vision is just 3-dimensional, for instance). A couple of weeks ago, as my feeling was that it was probably very high, I asked Turin directly for his guestimate on what the dimensionality of smell would be (besides time), and he said around 10 !!! So of course, the meaning of "shared" is fuzzier in a 10-dimensional space. And that is because the meaning of the underlying concept of "distance" is fuzzier in a 10D space, to the human mind anyways. And, yes, if you must know, I'm trying to map SMILES representations to general areas in the hypothetical odor space.
    Last edited by contrebande; 15th October 2020 at 02:35 AM. Reason: readability

  4. #4

    Default Re: Aldehydes, So-Called (SMILES)

    It's not the case that receptor fit (speaking or receptors in general) must be binary yes/no. In some cases rate theory is applicable, in others partial agonism, in other cases different biological effect due to cross-reactivities.

    The vibrational theory should not depend on inaccurate claims that if it's not accepted then one must accept binary fit, or any other misrepresentation of other possibilities. It should be able to stand on its own.

    It interested me for a while but not no longer does. If it interests you, great! But personally I would call it a stretch, no pun intended, to insist one finds similarities between decanal and hexyl cinnamic aldehyde, or the other examples I named. I do believe had you not been aware the latter is an aldehyde, you would never have said you perceived it as related, but I can't prove it of course, as you already have the information it is.

  5. #5

    Default Re: Aldehydes, So-Called (SMILES)

    Shape theory is binary, by its most generally accepted definition and I don't know about rate or shape/rate theory. I never said that smell theories were mutually exclusive either. In fact, I believe the exact opposite. The GPRCs are so big and the hypothetical spectrometer part so tiny within them that there very well might be other mechanisms involved on the same molecule to justify a syncretistic theory of smell in the end. That being said, you can call out my biases and/or call them what you want. You have yours, I have mine. And like you said, and as I've myself told you before, this forum is not the place to tell them apart. So here's yet another one for you (and posterity) : I believe AI will very soon come up with the best perfumes any which way a human smell consumer might evaluate them, consciously or not, autonomously or peer-pressured. With very little or very seemingly unrelated information about the target audience. It might already be the case. And those AIs (will) rely, again by definition, on euclidean odor spaces, however they are constructed. Machines might even come up with a 100% predictive and universal animal odor perception model before there is even a human peer-reviewed and generally accepted theory of smell to back it up. Biases or theory interests, yours or mine, don't even matter at this point. It's just data. And if there was no odor space to be derived from the data, perfumery itself couldn't have existed. It's a tautology : perfumery (like chess, like go, like art, like politics, like any "discipline") can exist if and only if an AI can eventually be made that beat humans at it. Furthermore, data and not smell theories is what motivated this post. I mentioned theories for illustration purposes. So, in the end, I agree with you : theories of smell are not interesting in themselves.

  6. #6
    Basenotes Junkie
    xii's Avatar
    Join Date
    Jun 2015
    Location
    Würzburg
    Posts
    971

    Default Re: Aldehydes, So-Called (SMILES)

    For instance ketones and SMILES.
    It suffices to parse the subtree containing the carbonyl group.

    A reduced SMILES string of an acyclic ketone should (if I didn't screw up) contain one of the strings below:

    RC(=O)R
    RC(R)=O
    O=C(R)R
    C(=O)(R)R
    C(R)(=O)R
    C(R)(R)=O

    where R is a placeholder having different properties in different positions. Once the string is found it should be easy to verify whether it's a ketone.

  7. #7

    Default Re: Aldehydes, So-Called (SMILES)

    Quote Originally Posted by contrebande View Post
    We are therefore in a continuous/stochastic realm instead of a binary fit/no fit one as in the shape theory of smell
    Quote Originally Posted by Bill Roberts View Post
    It's not the case that receptor fit (speaking or receptors in general) must be binary yes/no. In some cases rate theory is applicable, in others partial agonism, in other cases different biological effect due to cross-reactivities.

    The vibrational theory should not depend on inaccurate claims that if it's not accepted then one must accept binary fit, or any other misrepresentation of other possibilities.
    Quote Originally Posted by contrebande View Post
    Shape theory is binary, by its most generally accepted definition and I don't know about rate or shape/rate theory.:
    Well you are welcome to learn further on rate theory or partial agonism if you wish, or both. Nothing wrong with personally believing that receptor fit is yes/no, and have no knowledge of partial agonism or rate theory, and I have no wish to persuade you. But for others reading, merely again making an assertion "Shape theory is binary" goes nowhere towards demonstrating anything.

    For anyone else wishing to know: Shape of a molecule commonly does not merely switch on or fail to switch on a receptor, speaking of receptors generally. The final conformation of the receptor itself and its resultant activity can vary according to what it is binding. Secondly, in some cases the rate at which a ligand binds and unbinds can be important, not merely whether it is there or not (yes/no.)

    Not wishing to spend further time on molecular pharmacology, an example paper for anyone interested is https://sci-hub.se/10.0000/molpharm....ntent/36/3/437 , cited because I think it is the landmark paper in the field. If anyone wishes to take the time to research the subject and finds that this theory has been refuted since I left graduate school, please do, I have not kept up on rate theory since then -- in general, in medicinal chemistry which is my field it really does not matter or at least not in cases I know of. In contrast, partial agonism and cross-reactivity very much do matter. It would be a disaster in medicinal chemistry to adopt a mere yes/no as one's model. Seriously, that is at least 40 years out of date.

    Oh btw but likely very relevant to olfaction particularly in perfuming, I left out stimulus amplification... well outside of mere yes/no binary. Example, https://sci-hub.se/https://www.scien...207?via%3Dihub

    Leaping onto a given theory because one asserts that the more broadly or even nearly universally accepted alternate has some fundamental severe limitation (yes/no, merely binary) that it doesn't isn't always a correct leap.
    Last edited by Bill Roberts; 15th October 2020 at 02:35 PM.

  8. #8

    Default Re: Aldehydes, So-Called (SMILES)

    I see I missed answering your ketone question, replying on aldehydes.

    Why not think of a goodly number of molecules you know are ketones and not also aldehydes, alcohols, etc, and figure for yourself what you think may be reliable shared organoleptics of ketones?

    Then pick a goodly number of molecules where you know only the tradenames, and predict which are ketones and which are not from what you know of the smell.

    If comparing say Muscone with oh, say, Calone and various others has you picking up a "ketone" organoleptic commonality that lets you predict from smell whether molecules are ketones or not, where names don't give you a tipoff and you don't know the structure, that would be quite interesting indeed.

    That is the kind of thing that actually goes somewhere towards supporting a claim, and is do-it-yourself, see-it-yourself evidence as opposed to printed speculation that sounds clever (my characterization, fair or not, of vibrational theory.)

    It has to be before you know, as when it's after you know and I ask you if say vanillin, C10 Aldehyde, Hexyl Cinnamic Aldehyde, and Cortex Aldehyde all share an "aldehyde commonality" in organoleptics apparently your answer would be they do. Somehow no one had noticed this though and whatever that alleged organoleptic commonality would be if existing, it's useless to perfumery. Just saying.

  9. #9

    Default Re: Aldehydes, So-Called (SMILES)

    Hi Xii, the more I look into it, the more parsing the whole tree seems to be necessary. Even getting the "boundaries" of the subtree looks like it requires parsing the whole string first. I'm no longer trying to avoid it : basically now, I'm using CDK and looking at their SMILES tree object model (after the string has been parsed) to see if I can "search" the tree for "known" functional groups and index them separately, and them map them to organoleptic "tags" I can extract wherever I see them. I'm sure it has been done before by all the big corps who have been involved in proprietary AI perfumery, but it doesn't seem to have been published, yet.

    Bill, if a theory produces a computationally actionnable model, I will look into it. To me, in the case of perfumery, a "computationally actionnable model" is merely some kind of mapping of a molecule's objective/non-oranoleptic characteristics, properties and data to its subjective/organoleptic characteristics, properties and data. The vibration theory has done that, with the objective data here being the functional groups from the SMILES representation. Does the shape/rate theory provide a computationally actionnable model such as I have defined it ? If it does, I don't see it. I will not take a goodly number of ACs to make a goodly number of experiments. I will take a goodly number of datum to find a goodly number of organoleptic patterns, both from individual molecules and from a molecule's occurence in perfumery formulas.

    Quote Originally Posted by Bill Roberts View Post
    …whatever that alleged organoleptic commonality would be if existing, it's useless to perfumery. Just saying.
    Of course it very agressively "doesn't matter" to romantics in general and to luddites in particular, we all know that. I am neither and I say : perfumery IS organoleptic commonality.

  10. #10

    Default Re: Aldehydes, So-Called (SMILES)

    Quote Originally Posted by contrebande View Post
    Of course it very agressively "doesn't matter" to romantics in general and to luddites in particular, we all know that. I am neither and I say : perfumery IS organoleptic commonality.
    Don't get me wrong, I get where you're going... perfumery isn't magic and to treat it as such (and approach things with a lack of rigor) makes progress difficult.

    But by that same token, science isn't magic, either.

    What I mean by that is - if waving a chemical formula and speaking the magic words "organoleptic commonality" can't produce an actionable or measurable result in the real world, then you may as well be waving a magic wand and saying "abracadabra". Or, to be more topical, you may as well wave your hand delicately and say, "Trust your nose".

    Believe me, I've crashed into this wall over and over and over. I've come to brilliant scientific conclusions that make all the sense in the world on paper. I had a recent, particularly embarrassing conversation with Harry Sherwood where I presented my theory that esters don't work well in candles because they hydrolyze in the presence of the fatty acids from the wax. I was so proud of myself. I spoke so confidently. After all, I had the chemical models, I had worked out the science behind the reaction! I used fancy words like pKa and 'hydrolysis' and 'H+' and 'molarity'. He then informed me (in an exceptionally polite way) that I was dead-ass wrong.

    Reading this thread reminded me very vividly of that conversation.

    EDIT: Because it's both a topic of great interest and something I've studied for the better part of a decade, I can't help but chiming in on the below comment:

    Quote Originally Posted by contrebande View Post
    It's a tautology : perfumery (like chess, like go, like art, like politics, like any "discipline") can exist if and only if an AI can eventually be made that beat humans at it.
    Ermmm... Back in 1931, Kurt Gödel thoroughly disproved statements like the one in bold above.
    Last edited by NanashiSaito; 15th October 2020 at 05:02 PM. Reason: strange loops

  11. #11

    Default Re: Aldehydes, So-Called (SMILES)

    Quote Originally Posted by contrebande View Post
    We can agree on my disgust for magic thinking... and Secrets and exclusivity of knowledge is a recurring theme here
    Sounds like we're on the same page here. I too have noticed the trend towards secrecy, which as a hobbyist is vaguely annoying. But I definitely understand why someone who does this for a living would want to protect that, so I don't complain.

    Quote Originally Posted by contrebande View Post
    What did I say that you are referring to here ?.... I'm just trying to reproduce what is already being done by countless others behind closed doors, like IBM at Symrise.
    I'd say this part is what I'm referring to: the results you're shooting for. Until/unless your research actually yields something measurable, then it doesn't have any practical value. Or, at least not for the average Basenotes user. (Note that I'm not saying there's no value whatsoever. Just no immediate practical value. There may be abstract value, for sure, and believe me I love a good abstract discussion...)

    Quote Originally Posted by contrebande View Post
    In what way ? If there is empathy in what you are saying, I appreciate it. But I can't see the wall. And it doesn't feel as if I'm crashing into anything. Can you point out either to me with actual references to this thread ? Are you saying that I should be embarrassed in anyway ?
    Certainly not implying you should be embarrassed. Rather, trying to use my own embarrassment as an example of what happens when "this" gets taken too far. Not saying you have taken it too far, either. If I somehow caused offense, that wasn't my intent. Sorry! I acknowledge that I wasn't clear about what "this" is, so I'll clarify: when you have something that, on paper, should work one way, but in practice doesn't. I can't really speak to whether your own project has hit that wall. Hopefully you haven't and things are going well! I think that was the thrust of Bill's original question: it's unclear what results, if any, you have seen from your research. At least from what you've posted so far. I'd love to know more!


    Godel Turing Church etc.
    To be clear, I was specifically refuting the strong statement you had made (paraphrased): "any discipline can be mastered by an AI". I agree that Perfumery in specific can (and will eventually) be mastered by AI.

    The use of Godel to disprove that is certainly reminiscent of GEB, (recall the Tortoise besting Achilles' perfect record player by feeding it a record that causes the player to self-destruct). I don't think we need to fight about it, I mainly threw it in there as a bit of fun. And really, it's a purely academic, ridiculous digression on my part, which I don't think is necessary to expound on because it sounds like we are on precisely the same page. I have a deep and abiding interest in the more esoteric applications of AI... For example, I wrote a truly wretched piece of Harry Potter fanfiction in which the Singularity plays a major part of things.

    But to try to keep the conversation somewhat grounded, you have this statement that "AI will produce better perfume than humans". Which again, I agree with you (at least eventually). But you gotta consider your audience. One of my key functions in my day job is leveraging AI to... well, there's no nice way to put it... leveraging AI to replace humans. And it never really goes over well if I just march into a room full of programmers and say, "Heeeeyyyyy folks! I put together a Terminator-bot that's going to do your job for free!" Considering that several of the more active, helpful members of this community are perfume professionals, well, hopefully you can see how that statement might not go over well.

    I tend to find that the most productive, edifying reception of incoming AI technology comes when I frame things as "Here's how AI can help you do your job better" rather than "Here's how AI can do your job better than you." And there are oodles upon oodles upon oodles of ways that AI can help a perfuming professional. And from a purely logistical standpoint, there's going to be a very long time period where AI helps, before it's able to fully replace. So why not focus on that?

    At the end of the day, it seems like you have something valuable to contribute and I'd love to learn more, and include others in the discussion as well. So just trying to extend an olive branch of civility.
    Last edited by NanashiSaito; 16th October 2020 at 03:18 AM.

  12. #12

    Default Re: Aldehydes, So-Called (SMILES)



    Wowza!
    Last edited by foxbins; 16th October 2020 at 06:56 PM. Reason: Deleted quoted deleted post

  13. #13

    Default Re: Aldehydes, So-Called (SMILES)

    Quote Originally Posted by NanashiSaito View Post
    Wowza!
    To be fair, I didn't think so either. QED.

  14. #14
    Basenotes Junkie
    xii's Avatar
    Join Date
    Jun 2015
    Location
    Würzburg
    Posts
    971

    Default Re: Aldehydes, So-Called (SMILES)

    I we allow art to be rated, like perfumes by focus groups, we may indeed find ourselves beaten at it by a bunch of well programmed graphics processors. Rather soon. A way out would be to make art exclusive for humans to exercise. But, then again, having seen things like Anna Rudolf's analysis of AlphaZero performance against Stockfish we might observe warming to the idea of making AI integral part of humanity one day.

    I see Gödel's theorems made it to yet another internet discussion. The two being utterly unrelated, as usual.

  15. #15

    Default Re: Aldehydes, So-Called (SMILES)

    Quote Originally Posted by xii View Post
    I see Gödel's theorems made it to yet another internet discussion. The two being utterly unrelated, as usual.
    Whaaaaaaaaat? Perfumery and Incompleteness are totally related. If you'd like I can type up a 32-paragraph long screed wherein I tediously explain the connection using several questionable analogies and borderline non-sequiturs.

    (Since my previous attempts at good-natured tongue-in-cheek humor was met with a massive wall of vitriol, I'll clarify that I'm 100% kidding)

  16. #16
    Basenotes Junkie
    xii's Avatar
    Join Date
    Jun 2015
    Location
    Würzburg
    Posts
    971

    Default Re: Aldehydes, So-Called (SMILES)

    Quote Originally Posted by NanashiSaito View Post
    (Since my previous attempts at good-natured tongue-in-cheek humor was met with a massive wall of vitriol, I'll clarify that I'm 100% kidding)
    Since I was actually quite sure it was a joke, I'll clarify: I called it preemptively. In some boards mention of Gödel's theorems makes people grab meat cleavers. Good day.

  17. #17

    Default Re: Aldehydes, So-Called (SMILES)

    Quote Originally Posted by xii View Post
    Since I was actually quite sure it was a joke, I'll clarify: I called it preemptively. In some boards mention of Gödel's theorems makes people grab meat cleavers. Good day.
    Yep, very true. I was just having a laugh when I brought it up in the first place, thinking it would lead to some good-natured blathering. Sadly I was quite wrong! I forgot that Godel Escher Bach is sort of like the Atlas Shrugged of the esoteric academia world.

    I knew I should have gone with an Ayn Rand quote instead!

  18. #18
    Basenotes Junkie
    xii's Avatar
    Join Date
    Jun 2015
    Location
    Würzburg
    Posts
    971

    Default Re: Aldehydes, So-Called (SMILES)

    Not sure if relevant here but some annoying eristics might need to be addressed:
    Somewhat trained subjects presented with blotters dipped in short (straight chained) aldehydes and their respective alcohols are apparently able to tell one group from the other.

    Quote Originally Posted by NanashiSaito View Post
    At the end of the day, it seems like you have something valuable to contribute and I'd love to learn more, and include others in the discussion as well.
    My feeling exactly. I can't contribute much to the thread though.

  19. #19
    Wearing Perfume Right Now
    Bavard's Avatar
    Join Date
    Jul 2015
    Location
    Middlesex
    Posts
    3,165

    Default Re: Aldehydes, So-Called (SMILES)

    Some people have reported this thread for being upsetting. For anyone getting this feeling, it might help to avoid the thread altogether.

  20. #20
    Moderator


    Join Date
    Sep 2013
    Location
    NW US
    Posts
    3,985

    Default Re: Aldehydes, So-Called (SMILES)

    I have cleaned up this thread. If you can't discuss this issue without petty refutations of each other's views, I will close it. Please stay on topic, also.
    If you can't smell me from six feet away, I'm not wearing enough.

    It is the time you have wasted for your rose that makes your rose so important.--
    Antoine de Saint-Exupery.

    No baiting/trolling. No insults. No politics. Read the Code of Conduct.
    Currently wearing: La Nuit by Paco Rabanne

  21. #21

    Default Re: Aldehydes, So-Called (SMILES)

    Chacha Sikes whom I met through PK's FB group sent me something minutes ago I thought I must leave here too : https://www.aicrowd.com/challenges/learning-to-smell

    And what do you know, they are using SMILES representations to predict organoleptic tags. And Firmenich is sponsoring it. I'm just not hanging out in the right places it seems.

  22. #22
    Basenotes Junkie
    xii's Avatar
    Join Date
    Jun 2015
    Location
    Würzburg
    Posts
    971

    Default Re: Aldehydes, So-Called (SMILES)

    What about going the opposite way: there's a bunch of research about odour coding from the point of view of arrays of receptors. Notably by Harumi Saito and Joel D. Mainland. It maps (translates) molecules into receptor response charts. Which is enormous because receptor response is essentially additive. Once the data is sufficiently exhaustive, we'd be able to predict odours of mixtures just by knowing their components. But even with the existing information one could attempt to translate odour descriptors into receptor response by factoring objective (generated in vitro) receptor response through subjective odour descriptors from TGSC. Which could involve deep learning.

  23. #23

    Smile Re: Aldehydes, So-Called (SMILES)

    Hi Xii, thanks for your interest, my proposal here is to go for every possible way, not just accorging to just a single theory of smell. But so far, like I said, I have yet to find with the other theories of smell any computational pathway between the physicochemical properties of a molecule (or a mixture of stereoisomers as they are currently sold as a single "molecular" ingredient) and its organoleptic tags. For instance, in the papers that Bill Roberts suggested earlier on this thread, I saw nothing "actionable" in there for me. Do you know of other papers that could enlighten me on this? I need to find something that can be extracted or predicted out of the SMILES representation (like the functional groups, chirality, etc.) to map to the organoleptic tags. Like I said in my AI Crowd post, I am also looking at the GPCR data, but what I have access to seems very sparse in terms of coverage. What are the best, most exhaustive GPCR databases for olfactive receptors out there that are publicly available ? Does it specifically contains data for aromachemicals with CAS numbers so I can link them up to their TGSC profiles and their SMILES representations (that's the other problem I'm having with GPCR databases) ? I am also interested in predicted or empirical NMR spectra from the SIMLES representation (and there is a suggested way to do just that on TGSC). And I can go on forever about other intuitions I'm having on what other datapoints are relevant here.

    On the other end of the data, there are also a LOT of things to be done with the organolepic tags themselves (not just the SMILES string). Are they redundant ? Can they be transformed into a less sparse tensor/embeddings from a learned taxonomy (or an acyclic directed graph) instead of from simply a "sentence" mer of comma-separated organoleptic tags ? Can this taxonomy be augmented to a full ontology with learned properties organoleptic properties such as longevity and projection (and the likes). I used to be involved on Galois lattice connection theory for that, a long time ago, back in my ontological and NLP engineering days.

    I'm making a list of the different hypotheses that i think are worth to pursuing and I will post it on the AI Crowd forum and/or as a Colab Notebook (with code you can try yourself) and link to it here as well. And I might be in touch with you in private if I feel it exceeds what Basenotes DIY is capable of handling without too much catasrophic repercussions.

    Anyways, you know I'd greatly appreciate your input and even your coaching for this challenge, should you accept this mission !

  24. #24

    Lightbulb Re: Aldehydes, So-Called (SMILES)

    And I'm surveying into >100 Saito and Mainland papers just now on top of Firmenich's own Guillaume Godin's publications (another couple dozens there as well) and the python notebooks from the AI Crowd challenge. I'm not complaining, on the contrary. I can afford the luxury of focusing on that for as long as I need. I care less about participating in the challenge so much as broadening my horizons. And my, am I in for a broadening indeed !

    Also : SEND MORE PLEASE !!




Similar Threads

  1. Replies: 5
    Last Post: 11th April 2018, 07:52 PM
  2. When Ebay smiles upon thee....
    By BayKAT in forum Female Fragrance Discussion
    Replies: 8
    Last Post: 3rd January 2011, 07:21 PM

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •  



Loving perfume on the Internet since 2000