Reality #1:
LC/MS has become the primary analytical check for the medicinal chemist. I Love NMR, but it's the reality
So how did we get here? Clearly there a lot of reasons, but one of my hypotheses is based on the evolution of open access and the walk-up environment. No doubt that moving away from the traditional analytical service departments has been a great thing from a productivity standpoint.
LC-MS lends itself really well to the open access environment. There are good systems out there that can auto-process data well and provide a nice report that gives the chemist a simple answer. In most cases, the chemists can get the answer they are looking for with a quick glance at the report. In addition, for this kind of routine analysis, the interpretation of the results is very easy.
Contrast that to NMR. How often are the printouts from the instrument good enough? Not nearly as often. More and more it is becoming common practice for the chemist to import their raw or processed data into the processing software on the instrument or in some offline processing software. Within these applications, they can get a better look at their data, zoom in on areas of interest, re-integrate the spectrum so it makes sense, perhaps even generate a multiplet report for a potential patent down the road.
Add up those tasks alone, and the chemists may be forced to spend 5-10 more minutes on a routine NMR than an LC-MS when you factor in interpretation as well! Unfortunately, the NMR instrument is not printing out an answer, it is printing out a spectrum that needs to be interpreted by the chemists which on a routine basis is a lot more difficult and complex task than checking an LC-MS for reaction completion. Furthermore, let's keep in mind that intrepreting analytical data is not the medicinal chemists job. They are evaluated, compensated, etc. for making molecules. The more molecules the better so the inherent risk here is obvious and I don't think it's fair to simply "blame the chemists"
Given the above, is there any wonder why LC-MS has become the primary analytical check?
Reality #1:
LC/MS has become the primary analytical check for the medicinal chemist. I Love NMR, but it's the reality
Reality #2:
More often than not, NMR spectra are merely glanced at:
Reality #3:
In many organizations, upwards of 95% of compounds are being registered by synthetic/medicinal chemists without having undergone any analytical review by an analytical specialist (NMR spectroscopist).
***
How and why did we get here? I have some hypotheses, and will write about them in the coming days.
When I created this blog several years ago, one of the first links I added to my blog was to Stan's NMR Blog. Over the years I have had the pleasure to meet Stan who is a wonderful individual. His blog is very interesting, Stan is incredibly knowledgeable, and I really enjoy his writing style.
Recently, Stan posted about a topic, those readers who follow my blog know I am very passionate about; Automated Structure Verification (ASV). The initial entry was very thought-provoking and while I think the numbers Stan posts regarding the number of spectra chemists interpret below are too high, I think he's put together an interesting argument nonetheless about the need and application of ASV in the scientific community:
The typical application area for such a product is drug discovery, meaning Pharmas. I was told that these Companies employ masses of chemists who have to interpret some 50 spectra a day, having only 8h/50 = 9:36 minutes per spectrum. Clearly, they don't wag their tails, nor wish to pursue more balls. They were reduced to the echelon of routine workers and they hope that the new software might give them a few minutes more time to maybe ponder again that strange spectral peak that looked like an interesting challenge. Illusion, of course, because their top CEO's hope that the software will make half of the chemists redundant, leaving the rest doing a spectrum every 4:48 minutes (guess who will prevail).
Stan's entry definitely triggered some interesting comments from the community as well, in addition to a follow up letter from John Hollerton, GSK.
My take is that we don't need to be waiting 3-5 years for this technology to become more relevant...it already is! It is already being used in the industry. However, the secret is finding and recognizing the right application for it within the industry, and that application will vary considerably from lab to lab.
It seems the main discussion about this completely surrounds whether or not ASV can replace the NMR interpretation of the medicinal or synthetic chemist. Of course this is the ultimate goal, but it should not be considered the only goal. John eludes to some other applications in his letter and there are other uses as well.
I will blog more later about my thoughts on the "Reality of the New(ish) Discovery Environment" as it pertains to NMR and more thoroughly justify why I feel the way I do about this technology. But for me, the bottom line is that this is not just a concept, it's not a dream that we need to wait be realized in the next 3-5 years. It is also not a new technology on the horizon. Our first publication on this topic was back in 2006 and we've continued to build on those approaches in 2007 and 2008. Work was done by Griffiths et. al in the late 90s and early 2000s. The techonology has only gotten better and better. We have been working diligently with industry partners for over the last decade on this concept. It is not new and it is not an unsolved area.
Just to be clear, am I saying this technology is ready to completely replace the NMR interpretation responsibilities of the average medicinal/synthetic chemist? Definitely not.
Can it be used to help chemist make better, faster, and more independent decisions? Yes. Can it be used in complimentary way to the chemists existing interpretation workflow? Yes. Can it used to be validate the quality of registration libraries? Yes. Can it be used to verify compounds purchased from external sources, whether this are assay materials or simple starting materials? Yes. The list of applications go on for a while.
I believe current ASV applications can improve all the above workflows. The key is identifying the best applications and practicing caution in how this technology is deployed and used by chemists.
In conclusion, my advice is that if we continue to treat this technology as only applicable to the ultimate goal, perhaps it will take longer than 3-5 years, or perhaps NMR will become irrelevant before we get there. The goal of NMR scientists, hardware, and software vendors should be to work hard to continue to make NMR technology relevant in all industries. If we can do this, we all win.
I have blogged about Mikhail Elyashberg before and I wanted to take the opportunity to extend birthday wishes to him.
A tribute to Mikhail and his outstanding contributions in the area of Computer-Assisted Structure Elucidation (CASE) is published on the ACD/Labs website here.
At the age of 75 Mikhail is still working had in this area and you can stay tuned for a book on CASE coming soon. I will provide more details as they become available.
Congratulation Mikhail, and Happy Birthday!
Recently I finished a great book that made me think about different things on a lot of different levels. The book is called The Black Swan and the author is Nassim Nicholas Taleb.
In short, a black swan is defined as an event, positive or negative, that is deemed improbable yet causes massive consequences.
From wikipedia:
What we call here a Black Swan (and capitalize it) is an event with the following three attributes. First, it is an outlier, as it lies outside the realm of regular expectations, because nothing in the past can convincingly point to its possibility. Second, it carries an extreme impact. Third, in spite of its outlier status, human nature makes us concoct explanations for its occurrence after the fact, making it explainable and predictable.
Anyone who follows this blog knows that I have a passion around the concept of automated structure verification by software methods and the different applications it can play within the pharmaceutical industry.
Reading through this book, one of the things that popped in my mind was the black swan impact as it pertains to compound registration libraries in pharma.
I spend many hours in my role discussing the concepts of automated structure verification (by NMR of course) to spectroscopists, chemists, directors, VPs, Managers of compound management, etc.
In many of these conversations, the risk of having an incorrect compound in the registration library is always a fruitful and interesting discussion. However, this conversation is almost always dominated by low probabilities. The low probability that the compound being registered is wrong in the first place. I agree, this is a low probability. After all, a trained chemist has synthesized said material and has used a variety of analytical methods to confirm it's identity. In some organizations, the compounds in this library will be validated further either before they are accepted in the store, or before they are sent off to assay.
So the argument goes that incorrect compounds in the registration library are low probability. And furthermore, there are steps being taken in some organizations to proactively catch incorrect compounds (most often LC-MS). Finally, perhaps the lowest probability event of them all is the compound being identified as a hit during the assay.
So in the end, what are the chances that a meaningful compound that is identified as a "hit" is actually an incorrect compound? Furthermore, in this instance what's the chances that this compound will make it too far downstream without it's false identity being exposed?
In short, it's a pretty low probability event, and because in most cases it will eventually be caught it's a stretch to really call this a massive consequence or impact.
That's not the black swan I am talking about in this post.
In my opinion, the Black Swan concept is more relevant to those incorrect compounds that lie in the registration library and never get cherry picked out. A compound that doesn't produce anything interesting from the assay. A compound that effectively hides in the back of the shelves in a compound management cold room, that has been documented as "tested" but never advanced any further because of poor assay results.
Of course this compound is not the Black Swan. However, it's possible that the compound that it was supposed to be is. The compound that the chemist thought they made, and the compound that was referenced in the inventory.
While it's low probability, there are probably tens of thousands of misrepresented compounds that haven't truly been assayed over the last 20 years. Of these thousands could any of them turned into the highly coveted, blockbuster drug?
Sure, it's doutbful, but the infinitely small possibility exists.
And further, this very idea contains one of the core components of a Black Swan; "nothing in the past can convincingly point to it's possibility"
I am not certain that there is an example of a blockbuster drug that was originally missed because of a mistake in synthesis, or a misrepresented registrant. My guess is "not really" And if it did, perhaps as Taleb suggests, it would have been rationalized by hindsight.
Computer Assisted Structure Elucidation (CASE) is a major challenge in and of itself.
We've been working on this for many, many years and we continue to make huge strides in this area. The technology is sound, the software is being used successfully worldwide, and generally a month does not go by where I don't hear of a nice success story of how our software has helped a scientist elucidate a challenging unknown.
I recently had a good conversation with a customer/colleague who was complimenting us on our efforts in automated structure verification (more on our progress in this area, a little later). His comments went something like this:
"The work you are doing in this area is really impressive, but you do realize that this it today's problem, right?
"Please, please do not forget about tomorrow's problem, which is of course fully automated structure elucidation!"
We certainly haven't forgotten about that, but of course it is a challenging problem. And of course the major limitation might just be having access to the most information rich experiments that are available.
One thing is for certain with CASE. The same rules apply to a software as applies with human. If the data is of high quality and information rich, the problem is going to be much easier to solve.
During that discussion above, we happened to be working with scientists at DuPont, evaluating the impact of the 1,1-ADEQUATE experiment for CASE.
The 1,1-ADEQUATE has obvious benefits for interpretation either by a human or software. The challenging part about many of the traditional 2D experiments is that they are fuzzy by nature, meaning that the correlations are unambiguous (2, 3, or 4 bond). The 1,1-ADEQUATE provides unambiguous correlations which allows from a stronger filter. That said, it is most productive to combine information from a 1,1-ADEQUATE with some experiments that reveal ambiguous correlations to reveal longer distance correlations that aid in the process of structure elucidation.
In the study, we evaluated NMR data for two natural products and ran them through ACD/Structure Elucidator in different combinations.
The take home was that, not surprisingly, the 1,1 ADEQUATE experiment had massive impact on the accuracy and speed of structure generation.
For the first natural product, Retrorsine the following was observed:
HMBC alone: No structures generated after 16 hours of generation time
1,1 -ADEQUATE alone: 553 structures generated after 4 hours and 40 minutes. Retrorsine was the top ranked structure after generation.
1,1- ADEQUATE + HMBC: 2 structures in < 1 second
This publication revealed really an outstanding observation for the impact that the 1,1 ADEQUATE experiment can have on efficient CASE.
The article has been published in Magnetic Resonance in Chemistry and can be accessed here:
Enhanced Automated Structure Elucidation by Inclusion of Two-Bond Specific Data
Going back to the conversation I had, while this article certainly doesn't prove that fully automated structure elucidation is a reality, it does suggest one more step further in the ongoing process.
The key of course, is providing high quality, information rich experiments. However, before methods like this can really become applicable to a general audience these experiments have to become more accessible. Steve Cheatham and Michael Kline from DuPont did some really exceptional work here with some modern technology. NMR Experiments were run on a Bruker Avance 600MHz using a 5-mm TCI cryoprobe. 10 mg of sample yielded the acquisition times of 2.5 hours (16 scans) and 10 hours (64 scans).
Here's looking forward to the day where a chemist no longer asks, "Did I make what I thought I made?" and rather, "What did I make?"
I am delighted to announce ACD/NMR Processor Academic Edition!
You can download it for free here
We don't plan on making a ton of noise about this through our official website until next week some time, so subscribers to this blog and my friends in the blogosphere, you are the first to know, and please feel free to spread the word.
What is it? (If you are a subscriber to this blog, I hope you already know, but just in case)
This is a version of the software that is identical to the commercial version at the present time (with the exception that it includes freeware ChemSketch, instead of commercial ChemSketch). You can import 1D and 2D NMR data from all different formats, process them, perform multiplet analysis, create multiplet reports in patent and journal formats, automate procedures, create PDF reports, etc., etc., etc.
Over the last decade and more, this software has had a MAJOR impact on both academic and industrial environments. It has been an ideal fit since most organizations have made the transition to walk-up and open access laboratories. As an offline desktop processing package, you can install it on your laptops or desktop computers so that you can access, view, and process your lab reports, publications, or thesis from anywhere.
In this time we have also implemented some pretty sophisticated and cool things in the software for people who like to do more than just view, process, and report.
I could go on and on about all the things that this software can do.The best thing I can suggest is that you go and try it for yourself!
While doing that, please also visit, and subscribe to the new NMR Processor Blog
This blog will be run by my friends Joe and Andy, who will be updating with tips and tricks about the software (There's already a movie on Shortcut mode up there). Please contribute and comment as the more that happens the more the community benefits and the more the product can improve.
Anyway, I have been wanting to do something like this for a long time, and am thrilled that the moment has finally arrived!
Enjoy!
I was fortunate enough to attend the inaugural Structure 2010 conference in Leicestershire, UK last week.
This was a terrific event that focused on the structure elucidation of small molecules. The program for this conference was outstanding.
Unfortunately I didn't get the opportunity to attend every single talk, but the highlights for me were presentations by Lucio Frydman, Don Richards, and Christina Thiele.
Another great talk was given by my friend Mark O'Neil Johnson from Sequoia Sciences who truly is one of the best presenters I have ever had the privilege to watch. Great message, lots of pictures, and he's doing some really cool stuff.
I was also fortunate enough to do a workshop at this conference about Computer Assisted Structure Elucidation.
There was some overlap between my talk, Mark's talk, and Don Richard's talk and I think we all drew from the basic philosophy of the Faulkner Rules.
The Faulkner Rules were presented by the late great John Faulkner who was a professor at the Scripps Institution of Oceanography.
While quite straightforward and understandable, I think even after many years, the cardinal rules about elucidation still stand the test of time:
Rule 1: Never propose a structure before you have accumulated ALL possible spectral data. (Of course I believe these rules were first defined when 2D NMR was not all that routine and the abundance of different experiments were not available, but I still think this is a solid point).
Rule 2: If the structure is incompatible with any measurement, however minor, then the structure is wrong
Rule 3: Always find alternative structures before others do so and evaluate all alternative structures in a systematic manner.
Rules to live by, I think
David John Faulker 1942-2002
I've spent a lot of time over the past couple of years visiting both scientists and directors in Pharma and having discussions around the concept of automated structure verification, whether or not there is enough characterization being done on screens, libraries, purchased compounds, etc. Whether the possibility exists to do more characterizations and whether companies have the personnel to do this.
Derek's post entitled, the Infinitely Active Impurity doesn't necessarily capture all the feedback I have gotten, but it certainly does bring up some pretty solid arguments, and some of the comments are very interesting. Here are some excerpts from Derek's post:
Yesterday's post touched on something that all experienced drug discovery people have been through: the compound that works - until a new batch is made. Then it doesn't work so well. What to do? You have a fork in the road here: one route is labeled "Blame the Assay" and the other one is "Blame the Compound". Neither can be ruled out at first, but the second alternative is easier to check out, thanks to modern analytical chemistry. A clean (or at least identical) LC/MS, a good NMR, even (gasp!) elemental analysis - all these can reassure you that the compound itself hasn't changed.
But sometimes it has. In my experience, the biggest mistake is to not fully characterize the original batch, particularly if it's a purchased compound, or if it comes from the dusty recesses of the archive. You really, really want to do an analytical check on these things. Labels can be mistaken, purity can be overestimated, compounds can decompose.
And I've seen plenty of things that have fallen apart on storage, and several commercial compounds that were clean as could be, but whose identity had no relation to what was on their labels (or their invoices for payment, dang it all). Always check, and always do that first.
So why aren't these original batches fully characterized? Cost? Acquisition times? A human required to prep the sample? To interpret the results?
And some comments from Derek's loyal readership who share interesting stories, experiences, and practices:
In yet another program, we were doing an SAR study, and one particular compound didn't really fit the SAR the way we had predicted. Initially we took it in stride, and had all sorts of hand-waving arguments of why that compound might not fit the SAR. With time it became more obvious that something was rotten in Denmark - and a close look at the NMR spectrum revealed that in fact it wasn't the compound depicted, but a regioisomer.
Its probably less likely to occur these days I think due to modern tools and better upfront characterizations as suggested but I have heard of similar stories throughout my career and was involved in a several programs where the activity was from either from a minor by-product or a misidentified structure. The by-product was more puzzling at first but did lead to interesting series of analogs (before failing tox screen unfortunately). The misidentified compounds came from a wrong (mislabeled) regioisomer by supplier (so heed Derek's advice to pre-check purchased material as in this case was only obvious after a 2nd bottle gave a slightly different spectra) and another was an unanticipated chemical rearrangement during the synthesis (again only obvious after 2nd round gave less rearrangement). Although frustrating at the time these events were some of the most interesting times in the lab.
The other example came from a compound being contaminated in the Genevac. It's always a problem with shared equipment that many people don't think about. My own rule now on this is to always have LCMS and NMR (so many vendors only do LCMS) and if you get a good hit make a batch 2 on a bigger scale so you can really look at what's in there.
Anyhow, I've seen all of the above mentioned scenarios played out at some time or another. My sense is that the cost of a compound (i.e. how pure?, how sure?) should parallel the cost of the biology being done on it. You don't need EA, NMR, LC/MS, IR on each of 250,000 compounds going into a high throughput screen. And you better be mighty sure what you're putting into a dog tox study or into the clinic. If you get a high throughput hit, you need to confirm (more $) before you move forward very far (more $). There's nothing wrong with occasionally screening garbage if you don't act stupid after it hits.
My experience is that, at most companies, discovery research needs to spruce up its sloppy attitude to characterisation. (I dread to think what happens in academia, where we used to wait weeks for MS because the people were too busy doing *research* to be *bothered* with mundane structure confirmations). The approach that works in most cases has already been hinted at; synthesise batch 2, for preference by a different chemist, on a gram scale, and fully analyse (tlc and hplc) and characterise - Yes! Use C-nmr to check that all the carbons are there (!), and that their 1-bond and 3-bond couplings are logical. This is also an excellent nmr learning exercise for new BSc chemists.
The overall theme of this post, as well as the comments are that more diligent characterization is needed, even in high volumes.
This outlines one of the issues in this industry that we are trying to address through automated structure verification. But the argument for the system is of course that organizations do not have the personnel to diligently check the identity of the structure against analytical data from all different sources. An automated QC of these results can help address a good chunk of these concerns.
Of course one of the major concerns that comes up, is whether or not we can completley eliminate false positives. The answer of course right now is no. And of course it will likely always be no. Sometimes there are just really challenging problems.
Of course, nothing beats the diligent manual analysis by an analytical expert, but we all know that there is the time/money dilemma which is inevitable, and it's the reason why many organizations are unable to practice the due diligence that Derek and some of his readers are advocating.
In yesterday's post I pointed you to an article that has become the most accessed article of all time in the Journal of Cheminformatics, co-authored by representatives of ACD/Labs.
This article is a very comprehensive outline of the different approaches, developments, and successes of the development of CASE (Computer-Assisted Structure Elucidation) systems . I hope you agree that this technology has come a very, very, very long way, to the point that there are many spectroscopists around the world using ACD/Structure Elucidator routinely to identify unknowns in thier labs whether they be natural products, impurities, degradants, metabolites, etc.
Anna Codina from Pfizer is an example of just one of those users. She has been a long-time user of Structure Elucidator, and one of the very best in the world. She, along with her group, has really stretched the limits and applications of the software and has also been instrumental in the ongoing improvement of this software over the years.
At SMASH 2009, Anna shared some of her most recent experiences with the software in a presentation entitled, "From FID to Structure in 30 minutes?" This presentation can be accessed here.
For this presentation, she wanted to see whether it was possible to do a full elucidation starting from the raw data, all the way to the solution in less than 30 minutes. She was able to achieve this with several of her in-house Pfizer compounds, and then, for the purposes of this presentation reproduced these findings with compounds she could share in a pubic forum.In this presentation she goes through 3 such examples.
The most interesting piece for me was her discussion regarding the CASE limiting step in slides 9 and 10.
In an ideal world, we would be talking about completely automated structure elucidation. Of course, we are not quite there yet, but it is still a dream, as Elyashberg et. al refer to it in the J. Chem. Inf. article.
In the meantime, the process generally involves some manually handling of the data. For this process the general principle applies, "Garbage IN- Garbage OUT"
So the quality of data as well as the peak picking of said data is crucial.
The peak picking of multiple spectra is what Anna was referring to as the CASE-limiting step. We have made tremendous strides over the years to simplify this process as much as possible (Check out the demonstration movie here). But the traditional argument seems to still apply for some:
Why would I spend so much time peak picking all my 2D spectra?
By the time I finish picking and grooming the data, I have already solved the structure, so what's the point?
I will try to address these questions in a different post, but in the meantime, what's Anna's answer?
NMR Workbook, and specifically the NMRSync technology that we developed and released late last year. Note, that NMR Workbook is included with ACD/Structure Elucidator.
Anna proclaims that thanks to NMR Workbook, there is no longer a CASE-limiting step. With syncrhonized peak picking which carries through peak picking applied in one spectrum, to all others associated with the sample, the time spent to prepare the data is drastically reduced and is more consistent with the natural workflow.
For those unfamiliar with NMR Workbook and the NMRSync functionality, I introduced it on the blog about a year ago:
http://acdlabs.typepad.com/my_weblog/2008/12/if-all-of-these-spectra-are-from-the-same-compound-why-am-i-handling-them-all-separately.htmlIf you want to skip right to the demonstration go here:
http://www.acdlabs.com/products/spec_lab/exp_spectra/nmrworkbook/movie/NMRWorkbookFlash.htm
This product has turned out to be immensely popular and has evolved a long way in just one short year. Admittedly, it's probably been the biggest reason I have not been blogging as much over the past year, I have been touring the world showcasing this product, as well as our other solutions.
I hope over the next little while I can dive a bit more into the details of the product.Stay Tuned.