Skip to Content

False Positives Are Always Waiting

Here’s a paper in J. Med. Chem. that will have you adding another structure to the (long) list of Things to Watch Out For. The authors (from the University of Minnesota) report a screen that returned a series of furylquinolines as hits. Ones first reaction on seeing several structures from the same chemical series hit in a screen might be “Hey! SAR!” or (depending on how many times, or how recently) you’ve been burned, you might say “Hey! What do those things have wrong with them?”

In this case, what these structures have wrong with them is the furan ring. Furan is just not a group that shows up in medicinal chemistry very often, a notable exception in the horde of five-membered ring heterocycles that otherwise permeate the field. It has a reputation (earned) of being metabolically unstable, and as this example shows, it can be unstable to plain old ambient oxygen, much less the oxidative power of metabolic enzymes. The team found this out the hard way, retesting their own stock and then re-ordering commercial material, and finding these both repeating in the assay. But a freshly synthesized batch was inactive, and that’s when you squint at the assay results and go “Oh, crap”. Many of you can probably guess what’s coming next: as the freshly made material sat in DMSO solution, it magically began to pick up increasing activity over time. Ah, yes.

A closer look showed that the furan ring was doing a 4+2 cycloaddition with oxygen, and the resulting intermediate (basically a secondary ozonide) falls apart in various entertaining ways, including Baeyer-Villager chemistry. You get acids, aldehydes, lactams, and more, and some of these are likely Michael acceptors as well. The “SAR” was merely due to everything in the series falling apart into broadly similar piles of rubble. You’d have to figure that there are other substituted furans out there that are doing much the same thing in screening collections as we speak.

I’ve been there! And so has pretty much everyone else who’s had enough experience in dealing with primary screening data. This whole situation illustrates the fine line that you have to walk along. You need to be alert to the possibility of structure-activity relationships emerging from your data, of course (what kind of medicinal chemist are you if you aren’t?) But you also need to be alert to the possibility that you’re looking at false positives – because what kind of medicinal chemist are you if you aren’t? Accepting everything in the screening data at face value is a path to heartbreak and wasted effort, but the opposite approach (assuming that everything is crap) won’t get you anywhere, either. Just because you can screw things up in one direction doesn’t mean that you can’t screw them up in the opposite one – that’s one of the basic principles by which I conduct my scientific life – and the rest of my life, too, come to think of it.

When it comes to screening data, I do have a few structures that I rule right out (one beast that featured a quinone, a 1,4-diaminoaryl, and several free phenols all in the same structure comes to mind). Beyond that kind of thing, there’s a list of “Show Your Papers!” structures that I regard as “likely to be false positives”, and these furanylquinolines are now on it. These get labeled with stickers like “should probably work on other stuff first” and “make sure that the biologists don’t get too attached”. It should be fairly straightforward to show that these things are acting badly, if you’re alert to the possibility. (Resynthesis is one way, as with the paper under discussion, but the good old fashioned “run it through a plug of silica gel” can work wonders, too, and is usually much faster). Then come the “haven’t seen anything like that before, so I don’t know if it acts funny or not” compounds, which have to be dealt with impartially, and finally the “looks OK to me” set, which have to dealt with in the same way, because not everything that looks OK to me is, in the end, actually OK, dang it all.

It would be a lot easier and faster if we med-chem types could run our steely gaze down a long list of screening hits and say, with confidence, which ones (if any!) are the right things to spend time, money, and effort on. We can do that up to a point, of course, but that point isn’t nearly as far along the line as we like to think. Or like to have others think. . .

9 comments on “False Positives Are Always Waiting”

  1. John Wayne says:

    Most of us have had a lot of experience managing the expectations of others when it comes to potential screening leads. It is often a very frustrating experience telling other scientists that Compound 1 looks reasonable, while Compounds 2 and 3 are likely false positives; so, we’re going to establish that 1 is real, then look at 2 and 3.

    These 80/20 and 20/80 situations can be really hard to communicate to biologists/management/funding agencies because we have to admit that we’re playing an advanced guessing game based on decades of accumulated knowledge (both personal knowledge and the knowledge of the field as a whole).

    I once had a very senior manager stand in my lab with a reddish solution and a clear, colorless solution (same compound, the red one was from the screening deck).

    Manager: “Ok, so you are telling me that this compound (waves red solution) works, while this one (waves colorless solution) that you made doesn’t?”
    Chemist: “Yes.”
    Manager: “And you can’t tell me what this one is (waves red solution)?”
    Chemist: “Correct, they are spectroscopically indistinguishable. No fraction of the original sample is active; we tested them all.”
    Manager: “How are you going to address this?”

    At the time, I said something vaguely reassuring. In retrospect, I should have held my ground and said the the active species in the flask was probably a nonpatentable mixture of activated garbage that the FDA wouldn’t let us sell; our time is better spent working on something (or anything) else.

    If you buy me two beers, I’ll tell you what company I was with at the time.

  2. A Nonny Mouse says:

    Reminds me of the time when screening produced an array of varying structures that were all active in an important screen.

    After 6 months futile work making analogues with a large team, one said “isn’t it interesting that all of the active compounds are oxalic acid salts”- the biologists had failed to mention this point to the chemists.

  3. David Borhani says:

    Fascinating and very informative publication!

    I don’t know the literature of singlet oxygen photosensitization at all, beyond Rose Bengal and a few other beautifully colored dyes. Does anyone who does know the literature understand the (to me, apparently idiosyncratic) SAR in the mechanistic analogues shown in Table 1? Presumably it has much to do with either the absorption & fluorescence spectra and excited state lifetimes of each compound, or (again, seemingly idiosyncratic) sensitivity/reactivity with singlet oxygen, or both…

  4. DCRogers says:

    This overlaps with issues of promiscuity – one modeling adventure used Bayesian with ECFPs and massive amounts of historical data from a wide range of campaigns.

    We treated as ‘active’ a compound being a hit in some number N (or more) screens. Analysis of the features preferred by the model indeed found some features known for on-target promiscuity; but a large number of features were associated with false-positives, that is, features that interfere with the assay without actually having any on-target effect.

    In our case, either effect was unwanted, and so the model was useful to flag questionable hits in newly-performed screens, hopefully helping to tamp-down early over-optimism.

  5. anon the II says:

    I haven’t commented in a while so here goes.

    John Wayne does a good job of describing a scene that played out repeatedly in every pharma company that did high-throughput screening. The problem is that, back in the day, when an experienced medicinal chemist said “that don’t look like a drug”, his experience was valued and his advice was heeded. Over the last 20 years, that’s gone.

    Back in the comby days, a lot of libraries were made by mixing an amine and an aldehyde and then doing some chemistry. Unfortunately, those intermediate imines could be quite stable and if left in the stew, they would tend to light up assays which invariably led to the political issues Dr. Wayne described. We tried, without much success, to argue that if the activity didn’t make it through a C18 column, we should drop it and move on.

    If those amines or aldehydes used to make libraries were furans like aminomethylfuran or furfuraldehyde, you could always tell which row or column they were in because it was brown to black. If you used both, you got a nice black cross of wells in an otherwise clear 96-well plate. We called it the “furan cross” to give our work a religious flavor.

    Lapatinib has a furan. Making derivatives for an affinity resin or adding fluorescent tags proved to be a miserable experience. Once a linker rendered it non-crystalline, it slowly but surely degrades into a plethora of products. Try explaining that to an MD.

    Finally, my initial reaction to this paper was “Why bother?” But recently, I tried to explain to someone that had a hit with two furans in the molecule, that it wasn’t worth following up. He wasn’t convinced. Now I have a paper to point to.

  6. milkshake says:

    re furans: you need one electron withdrawing substituent, carboxamide, nitrile, or at least aryl, otherwise the compounds are too metabolically unstable, and unpleasant to work with. The same goes with pyrroles. Indoles are a pain in the ass too. Every time you have highly electron rich aromatic you ask for trouble, and if the spot turns brown on TLC after looking at it under UV lamp, it will probably photooxidize in solution on sunlight too.

  7. RTL says:

    We recently characterized a lesser-known PAINS compound that looked really interesting, until we confirmed that it covalently bound to our target in a manner that was not so interesting for drug development: http://www.ncbi.nlm.nih.gov/pubmed/26296926

    The list of compounds with latent, non-druglike characteristics is getting long. Fortunately, with a skeptical eye, risk usually can be discharged via a handful of straightforward biochemical assays. As Mad-eye Moody says: “Constant Vigilance!”

  8. Toluene says:

    Does anyone like benzo[c]furans as a developable template?

  9. Chrjgra says:

    Although I’m woefully unqualified to comment (ex-drug discovery biologist) your article rings a bell from when I was running a blue sky, high risk psychiatry project at one of the european pharmas with a long history in antipsychotics during the heyday of high throughput screening – I remember my medchem co-project leader and his boss gently explaining some of the issues you talk about and in the associated comments and hope I learned a little bit about medicinal chemistry – I don’t recall how many thousands of compounds from combichem and bought in libraries of perhaps dubious quality we tested in the primary functional screens but i can still see the 96- and 384-well plates being loaded on to the FLIPR and getting highly excited at the apparently positive responses from the HEK cells my team had genetically engineered, only to learn that they were likely false (some of them were also highly coloured) – thankfully we did eventually find one or two real hits and these went on into development, at least as far as Phase 2, which was gratifying after having to putting up with a lot of the politics and naysaying from senior management you refer to above.

Comments are closed.