RT @curiouswavefn: @KRHornberger I think @pwk2013 in particular would appreciate that. Asking to look at the raw data is what he did in his…
@KRHornberger I think @pwk2013 in particular would appreciate that. Asking to look at the raw data is what he did in his famous paper https://t.co/MQcs7nZiNX
RT @pwk2013: @olexandr I have found that people don't always like it when I cite their papers 🤣🤣🤣 https://t.co/2SKG7az2Xr
@olexandr I have found that people don't always like it when I cite their papers 🤣🤣🤣 https://t.co/2SKG7az2Xr
@KRHornberger I think that you need to purge yourself of all this negativity although, if pressed, I'd have to admit to writing "the human liver remains an effective antidote to the hubris of the drug designer" in the notorious correlation inflation articl
RT @pwk2013: @KRHornberger "Given that drug discovery would appear to be anything but simple, the simplicity of a drug-likeness model could…
@KRHornberger "Given that drug discovery would appear to be anything but simple, the simplicity of a drug-likeness model could actually be taken as evidence for its irrelevance to drug discovery." https://t.co/2SKG7az2Xr
RT @pwk2013: Correlation inflation: http://t.co/vDve6w57
It's been a decade (and 3 days) since the debut on twitter of 'Inflation of correlation in the pursuit of drug-likeness' #cheminformatics #DataScience #MedChem #PourEncouragerLesAutres
@DrBostrom @andrewwhite01 @wpwalters SFI is neither optimized nor validated for prediction of solubility (weighting between logD and #Ar is arbitrary and was not determined by the data). My advice is to be very wary of analyses in which continuous data are
@BZdrazil "Although data analysts frequently tout the statistical significance of the trends that their analysis has revealed, weak trends can be statistically significant without being remotely interesting." https://t.co/2SKG7az2Xr
@curiouswavefn @philipcball @RommieAmaro @A_Aspuru_Guzik I did consider describing Ro5 as a 'sacred cow' in the correlation inflation article (as I have in blog posts). After thinking long and hard about it, I concluded 'maybe not' 😁😁😁 https://t.co/2SKG7a
@5_utr @gdgrass Oh dear!!! Nice example of correlation inflation: https://t.co/nRXAs2xZ0q Thankfully my taxes/charity donations dont feed that sort of research. What a relief!
@HewittPete @covid_moonshot The position of the 3D-ness evangelists is compromised by some rather dodgy data analysis as detailed in this heretical study (we were lucky to avoid the auto-da-fé) https://t.co/2SKG7az2Xr
@aminoacidhead The challenge is to make chemical space and drug-likeness (screening concepts) relevant to drug design (a post-screening activity). Shaky data analysis compromises credibility of drug-likeness concept: https://t.co/2SKG7az2Xr
@HartungIngo You might want to check the sp3 carbon count for these structures. Some of the published sermons on the goodness of sp3 carbons are based on shaky data analysis: https://t.co/2SKG7az2Xr
@KRHornberger @drug_smolecules Bedtime for me here in UK although I should thank the "property inflation" crowd for providing inspiration (not least for the title) for this article: https://t.co/2SKG7az2Xr
@SuperScienceGrl Here are some practical suggestions for making the correlations more visible: https://t.co/2SKG7az2Xr
@joaquinbarroso @itamblyn From Dirac, it's quaint. From drug discovery scientists, it's pretentious. I recall encountering the term "physics envy" in a book by Steven Jay Gould and we introduced the term "Ro5 envy" in this article: https://t.co/2SKG7az2Xr
"Although difficult to define objectively, the characteristic of a lead series that is most prized by medicinal chemists is room to maneuver." #MedChem #DrugDesign #LeadOpt #cheminformatics https://t.co/2SKG7az2Xr
RT @pwk2013: @jonaslindeloev 'Inflation of correlation in the pursuit of drug-likeness' (see also 'voodoo correlation') may be of interest…
@jonaslindeloev 'Inflation of correlation in the pursuit of drug-likeness' (see also 'voodoo correlation') may be of interest https://t.co/2SKG7az2Xr
@zenbrainest @jeremymberg argh, please no binning https://t.co/e3Y70fol3V
@fxcoudert Maybe it was a presentation on druglikeness https://t.co/2SKG7az2Xr
RT @pwk2013: "Not all #attrition is equal and stopping a project after a Phase I study will always be more palatable than being forced to d…
"Not all #attrition is equal and stopping a project after a Phase I study will always be more palatable than being forced to do so by #toxicity during a Phase III study." #MedChem #DrugDesign #DrugDevelopment #ClinicalDevelopment #ADMET #DMPK https://t.co
@daniela_witten "Weak trends can be beaten and when powered by enough data, even the feeblest of trends acquires statistical significance" https://t.co/2SKG7az2Xr
RT @pwk2013: "Prediction is very difficult, Niels Bohr is said to have observed, especially about the future and the human liver remains an…
RT @pwk2013: "Prediction is very difficult, Niels Bohr is said to have observed, especially about the future and the human liver remains an…
"Prediction is very difficult, Niels Bohr is said to have observed, especially about the future and the human liver remains an effective antidote to the hubris of the drug designer." #MedChem #DrugDesign #cheminformatics #DMPK #ADME https://t.co/2SKG7az2Xr
“...those who have followed the #DrugDiscovery literature over the last decade or so will have become aware of a publication genre that can be described as ‘retrospective data analysis of large proprietary data sets’ or, more succinctly, as '#Ro5 envy'.” h
#Druglikeness: the branch of #MedChem in which the beauty of the molecular structure is more important than the in vivo behavior of the compound #cheminformatics #DrugDesign https://t.co/2SKG7az2Xr
@HartungIngo @JMedChem @NovartisScience Strengths of relationships between #ADMET properties (e.g. aqueous solubility & permeability) and physicochemical properties (e.g. #logP/#logD & molecular size) have been 'talked up' for years. #druglikeness
@vijaypande Don't knock binning! It's a great way to massage trends in data 😁😁😁 https://t.co/2SKG7az2Xr
@FuchterMatt @JMedChem @impchemistry When blazing new trails it helps to be aware that some of the perceived #Medchem 'wisdom' (e.g. benefits of '3D-ness') is based on shaky analysis and presentation of data. This article discusses correlation inflation (
Wannabe 'thought leaders' in property-based #DrugDesign may appreciate these practical suggestions on how to invigorate anemic trends in data so that you can better lead the thoughts of the Great Unwashed #cheminformatics #logP #druglikeness #DataScience
@SygnatureDiscov @RSC_BMCS @RSC_CICAG Inflation of correlation in the pursuit of #druglikeness may be of interest although I'm guessing that the speaker won't be referring to it https://t.co/2SKG7az2Xr
@JeremyGreenwoo4 @macinchem @curiouswavefn Generally binning of continuous data is done to make trends look stronger than they actually are. If you're using continuous data to train models, you need to use RMSE to validate. This article may be relevant: h
@gretchen_H20 @callin_bull @UMNFWCB @UMNIonE @UMNews I like it! There is a considerable amount of BS in drug discovery and this article highlights some crude (although highly cited) examples of correlation inflation (also known as voodoo correlation) https
@biogerontology I am particularly skeptical about claims that chemical matter is drug-like. The basis of drug-likeness is weak and some rather questionable data analysis has been published in the literature as detailed in this article: https://t.co/2SKG7az
@MedChemica It might be worth mentioning that the article provides a textbook example of correlation inflation (see https://t.co/2SKG7az2Xr )
@codehovel @curiouswavefn That is the basis of #druglikeness 😀😀😀https://t.co/2SKG7az2Xr
@RSC_BMCS @SygnatureDiscov @RSC_CICAG @RoySocChem @Dereklowe @EuroMedChem @MedChemComm @JMedChem @NatRevDrugDisc You can't actually say since only drugs (defined as compounds that had been taken into phase 2 trials before the original publication) included
@sengupso @curiouswavefn @jenheemstra In #DrugDiscovery both strength and relevance of trends in data tend to get talked up and the dearth of public debate points to a malaise in the discipline. This article may be relevant to the discussion: https://t.co
@SygnatureDiscov You might want to examine the #HydrogenBond donors and acceptors in the #ibupfoen molecule a bit more carefully. The comments on #Ro5 in the introduction of the following article may also be helpful https://t.co/2SKG7az2Xr
@DDU_Dundee @lawebster10 #Ro5 is simply a statement of property distributions for compounds that had been taken into #Phase2 before publication of the rule in 1997. This 2013 article provides some critical review of the rule: https://t.co/2SKG7az2Xr
@ElisaTelisa @MULibraryCat @Claire_McGing Really should have used a Hannibal Lecter reference for "... the human liver remains an effective antidote to the hubris of the drug designer" https://t.co/2SKG7az2Xr
@jaybradner @OlallaLab @NatureNews @Scudellari Some of the #druglikeness dogma is based on flaky analysis of proprietary data and the link between physicochemical properties and outcome is not as strong as the 'experts' would have us believe #cheminformati
@dud3k @NatRevDrugDisc @Dereklowe Worth remembering that #Ro5 is simply a statement of property distributions for compounds that had been taken into #phase2 studies before the article was published in 1997. More detailed discussion about Ro5 in this articl
@LindorffLarsen You may find 'Inflation of correlation in the pursuit of drug-likeness' to be relevant (and possibly even droll at times) https://t.co/2SKG7az2Xr
@HCCvPDAC @davidlmobley @MichaelKGilson Need to select your targets very carefully. In #DrugDiscovery the targets are often linked to each other for mutual reinforcement which makes them easier to hit (as was the case for our correlation inflation study)
@curiouswavefn @rguha @Dereklowe "...those who have followed the drug discovery literature over the last decade or so will have become aware of a publication genre that can be described as ‘retrospective data analysis of large proprietary data sets’ or, mo
RT @pwk2013: @HartungIngo @CarolynBertozzi @NovartisScience @JMedChem We noted "Given that drug discovery would appear to be anything but s…
@HartungIngo @CarolynBertozzi @NovartisScience @JMedChem We noted "Given that drug discovery would appear to be anything but simple, the simplicity of a drug-likeness model could actually be taken as evidence for its irrelevance to drug discovery" in http
@cguibourg Here are some tips on squeezing correlations out of refractory data sets: https://t.co/2SKG7az2Xr
@adrian_roitberg @curiouswavefn A difficulty I experienced in my correlation inflation study ( https://t.co/2SKG7az2Xr ) was the proprietary nature of most data (except for solubility/Fsp3) were proprietary. Current @JMedChem author guidelines inhibit data
@chupvl @rguha And the thresholds can be manipulated to influence our decisions (and even our thinking) https://t.co/2SKG7az2Xr
@hannahbruce Although I don't pay much attention to citations of JIF to assess articles, highly cited articles provide excellent cannon fodder for critiques like this one: https://t.co/2SKG7az2Xr
@curiouswavefn @rguha @JohnDelaney8 @Dereklowe It's worth looking at the analysis that was presented in support #PFI when it used to be called solubility forecast index #SFI (see comments on that analysis in https://t.co/2SKG7az2Xr)
@rguha @davidlmobley @jchodera @curiouswavefn Given that drug discovery would appear to be anything but simple, the simplicity of a drug-likeness model could actually be taken as evidence for its irrelevance to drug discovery https://t.co/2SKG7az2Xr
@Publons @ChemConnector Just don't mention the error bars... https://t.co/2SKG7az2Xr
@OrkidDr Here are some practical tips that may be helpful if you really want to get them spitting feathers https://t.co/TJgPvwmlc1
@jeanne_erdmann @openscience @RachelZamzow @Open_Notebook This may also be of interest https://t.co/TJgPvwmlc1
@leecronin I'll see what I can do but in the meantime here are some tips for making alternative facts seem less alternative https://t.co/TJgPvwmlc1
@JohnCMohl @phylogenomics Correlation inflation is also quite scary https://t.co/TJgPvwmlc1
@nathanbroon @dgelemi @indianpaddy Comments on #Ro5 in correlation inflation article may be helpful https://t.co/2SKG7az2Xr
@Jrvanalstine @curiouswavefn @RoundSqrCupola @sundarpichai Some of us subscribe to the idea that "sacred cows make great hamburger" https://t.co/TJgPvwmlc1
@exnovodiscovery @alxndrkalinin @jchodera @rbhar90 @vijaypande Some effort #DrugDiscovery is devoted to making weak trends look less so... https://t.co/OHvuIVzm21
@LucyRForrest @Sci_Lipi Use of 'creative' in data analysis context is not always complimentary and this may be of interest: https://t.co/OHvuIVzm21
@kyleserikawa @curiouswavefn @Dereklowe Bear in mind that data may be presented in ways that exaggerate trends within it https://t.co/OHvuIVzm21
@jchodera @rbhar90 A related problem is that trends in data can be made to appear stronger than they actually are: https://t.co/OHvuIVzm21
@TahaYasseri @AdrianMulholla1 Absolutely but you still need to analyse the data https://t.co/OHvuIVzm21
Practical guide to #correlation improvement (re-post) #cheminformatics #MedChem #druglike #CompoundQuality #JCAMD https://t.co/OHvuIVQWTz
@CT_Bergstrom Here's some material that may be helpful (see also voodoo correlations): https://t.co/OHvuIVzm21
From 4 years ago #druglikeness #MedChem #cheminformatics #SmokeAndMirrors https://t.co/OHvuIVzm21
@EnricoCoiera @Dereklowe The #druglikeness field appears to be particularly blighted (see also #VoodooCorelations ): https://t.co/OHvuIVQWTz
RT @GaryMirams: An interesting paper on how to avoid/spot artefacts of data processing inflating #CorrelationCoefficients https://t.co/gs1x…
An interesting paper on how to avoid/spot artefacts of data processing inflating #CorrelationCoefficients https://t.co/gs1xRQZZ5P #Stats https://t.co/eqHsu0uGPz
Author of https://t.co/soG87DvkhI needed to read https://t.co/OHvuIVzm21 before pontificating on state of #MedChem cc @JMedChem
@stuartcantrill Would you have dared to publish this? https://t.co/OHvuIVzm21
@Brooke_LaFlamme @HawaiiSci Original publications are not safe even when authors hide the data: https://t.co/OHvuIVzm21
Worry about logP & MW if ADMET assay results => acceptable compound profile? #MedChem #cheminformatics #CompChem https://t.co/OHvuIVzm21
@DrBostrom @curiouswavefn @edsherer @Dereklowe @Chemjobber @MatToddChem Would this have satisfied publishing goals? https://t.co/OHvuIVzm21
@dgelemi A weak correlation tells us that effect only explains small proportion of variance https://t.co/OHvuIVzm21
@WendyAnneWarr Correlation is not yes/no so statements like this are not meaningful? See also https://t.co/OHvuIVzm21
#DrugAction driven by affinity & unbound concentration (not #LigandEfficiency & unbound fraction) #MedChem #PhysChem https://t.co/OHvuIVzm21
@matthewherper #Druglikeness and #CompoundQuality experts gush about #StatisticalSignificance but ignore #EffectSize https://t.co/OHvuIVzm21
@davidlmobley @curiouswavefn Octanol/water logP doesn't tend to 'see' hydrogen bond donors https://t.co/OHvuIVzm21 https://t.co/LI3zun7tEW
@NatureNews @jchodera @nature See slide 11 https://t.co/4aav6WEaVJ and https://t.co/OHvuIVQWTz for more on correlation inflation problem
@fraser_lab @pollyp1 @mbeisen Maybe and you can also do #PPPR this way https://t.co/OHvuIVQWTz
@EvoMRI @Science_Open @openscience I also do post-pub #peerreview in journal articles https://t.co/OHvuIVzm21 https://t.co/DjbFlgbnsN
@curiouswavefn @stuartcantrill @DevilleSy @MCeeP Some bloggers also publish... https://t.co/DjbFlgbnsN https://t.co/OHvuIVzm21
@JCIM_ACS @kmerz1 Averaging data prior to analysis also works well https://t.co/OHvuIVzm21
@curiouswavefn @Chemjobber @Dereklowe @michael_gilman Symptoms of the malaise? https://t.co/OHvuIVzm21 https://t.co/DjbFlgbnsN
@all_isee If so, it's worth noting that some Fsp3 and aromatic ring count studies are a little flakey (see https://t.co/OHvuIVzm21 ) (4/N)
@Sci_Lipi And a vital first step! Hazards of substituting visualization for analysis illustrated in: https://t.co/OHvuIVzm21
RT @pwk2013: Hopefully not how #FutureTimeChem will turn out https://t.co/m5dwpHji5f https://t.co/DjbFlgbnsN https://t.co/OHvuIVzm21 #Real…
Hopefully not how #FutureTimeChem will turn out https://t.co/m5dwpHji5f https://t.co/DjbFlgbnsN https://t.co/OHvuIVzm21 #RealTimeChem