@alex_mackerell [3/3] Bear in mind that a protein-ligand contact can be destabilizing without being inherently repulsive and the contribution of protein-ligand contact to affinity is not an experimental observable (the second point is discussed in: https:/
@KRHornberger In NoLE I observed "... the most interesting SAR is likely to be associated with the most deviant values" (the somewhat obscure wording was quite deliberate and celebrates the antics 🐷🐷🐷of Parliament's Eton-Oxford mafia) https://t.co/2xu214
@carolvorders I've always found it rather droll that naughty Tories are punished by not being whipped and I wrote that "the most interesting SAR [a piece of drug design jargon] is likely to be associated with the most deviant values" with naughty Tories in
@KRHornberger There are other problems: "Sample bias can be significant, even in large datasets, as exemplified by divergent conclusions of two apparently similar studies [41, 46] with respect to the relationship between pharmacological promiscuity and mol
@rflameiro @JCIM_JCTC @CalleaLara @andrrizzi @devivo_marco @Paolo__Carloni My comment was more about the objective of the design than assessing likelihood of success. Need to think beyond affinity if calling it #DrugDesign and second paragraph of intro to
@JCIM_JCTC When decomposing values of ΔG° it's important to aware that ΔG° varies with the standard concentration. See 'Thermodynamic aspects of ligand–protein association' section in NoLE: https://t.co/2xu214v5cB
@raw883 @HartungIngo @D_B_McConnell I don't agree. Contribution of an intermolecular contact to binding affinity of a reversibly-bound ligand is not an experimental observable. 'Thermodynamic aspects of ligand–protein association' section in NoLE may be re
@BShoichet @D_B_McConnell @KRHornberger The ‘Thermodynamic aspects of ligand–protein association’ section in my #LigandEfficiency critique is relevant to this discussion. [5/5] https://t.co/2xu214v5cB
@KRHornberger @med_chemist I offer this advice in Nature of LE: "Drug designers should not automatically assume that conclusions drawn from analysis of large, structurally-diverse data sets are necessarily relevant to the specific drug design projects on
@EnamineLtd If used with Ro5 HB acceptor definitions, Ro3 will eliminate carboxylate bioisoteres like tetrazole as well as a number of anionic groups. See comments on Ro3 in 'Ligand efficiency and fragment-based design' section of this article: https://t.c
@HartungIngo @ACSBioMed Bear in mind that relative LE values for initial fragment hit and final lead vary with concentration unit used to express affinity (or potency). Discussed in 'Ligand efficiency and fragment-based design' section in 'The nature of li
@prof_ajay_jain @AnnJainPhD @edsherer @BruecknerAlex @mreibarkh #LigandEfficiency is physically meaningless and has no thermodynamic basis because perception of efficiency changes when you use a different concentration unit to express affinity. See Table 1
@D_B_McConnell It can be useful to include controllability of exposure as one of the objectives of drug design (see 2nd paragraph of Nature of Ligand Efficiency; if I was writing it today I'd mention free drug hypothesis) https://t.co/2xu214vD29
@callin_bull I suspect that the physically meaningless (perception changes when you express a quantity in a different unit) ligand efficiency metric might be the type of thing you're talking about: https://t.co/2xu214v5cB
Promote it on @nftssaga
When claiming that an #ML/#AI model is interpretable, it is important to be aware that the contribution of a structural feature to affinity is not an experimental observable #cheminformatics #CompChem https://t.co/2xu214v5cB
RT @pwk2013: Although widely used, #LigandEfficiency is an unsuitable #metric for #DrugDesign because perception of efficiency varies with…
RT @pwk2013: Although widely used, #LigandEfficiency is an unsuitable #metric for #DrugDesign because perception of efficiency varies with…
Although widely used, #LigandEfficiency is an unsuitable #metric for #DrugDesign because perception of efficiency varies with the unit used to express affinity (or potency) and I’ll re-post this 2019 #OpenAccess article #MedChem #cheminformatics https://
Difficulties associated with writing ΔG° as a sum of contributions by individual interactions become obvious by trying to write K.d as a product of analogous contributions (see "It has been asserted [54] that..." in Nature of #LigandEfficiency) https://t.
@ChemicalBiology @D_B_McConnell I would argue that contributions of individual interactions to affinity, selectivity or kinetics are not experimental observables. 'Thermodynamic aspects of ligand–protein association' section in this article may be relevant
RT @pwk2013: @PracticalFrag @GastreichM @biosolveit @EnamineLtd @ACSPublications In fragment elaboration, consider modelling response of pI…
@PracticalFrag @GastreichM @biosolveit @EnamineLtd @ACSPublications In fragment elaboration, consider modelling response of pIC50 to molecular size (HA; MW) and/or logP, and using residuals to quantifying efficiency as discussed in 'Alternatives to ligand
@julienmich80 @hannahbruce @D_B_McConnell @jchodera @davidlmobley @zoecournia @JorgensenWL @chem_essexgroup @DennisWhom Absolutely! I highlighted (including citation of your excellent study) non-additivity as an issue when discussing group efficiency (ΔΔG
@hannahbruce @D_B_McConnell @jchodera @davidlmobley @zoecournia @JorgensenWL @chem_essexgroup @DennisWhom This difficulty is discussed (in excruciating detail) in ‘Thermodynamic aspects of ligand–protein association’ section of ‘Nature of ligand efficiency
@D_B_McConnell The dependence of ΔS° on C° is determined by stoichiometry (see comments on ref 54 in thermodynamics section of Nature of Ligand Efficiency) (3/3) https://t.co/2xu214v5cB
@D_B_McConnell For many equilibria (including target-ligand binding) TΔS° depends on standard state definition and any change in perception resulting from using different standard state definition is not meaningful (discussed in thermodynamics section of
@RenzoCarlucci @D_B_McConnell I use it conceptually although it may be possible to define & quantify it computationally. For an example, see "Target interaction potential (TIP) can be a helpful concept..." in Nature of Ligand Efficiency: https://t.co/2
@D_B_McConnell @BShoichet @MichaelKGilson MAoL is a favorite of mine and a true classic. However, intercept appears to have been forced through Kd = 1 M and it's worth pondering likelihood of observing maximal affinity with data set of that size (see MAoL
@ilpeppe It is not well known that I included "the most interesting SAR is likely to be associated with the most deviant values" in Nature of LE as a homage to England's ruling class https://t.co/2xu214dtO1
RT @pwk2013: @MedChemica The belief that one can infer the importance of individual interactions from inspection of the structure of target…
@MedChemica The belief that one can infer the importance of individual interactions from inspection of the structure of target-ligand complex is a delusion. Contribution of an intermolecular contact to binding affinity is not, in general, an experimental o
RT @pwk2013: @QuantumTessera @radscientist_ Gutted to have missed it! I think it's over 10 years since I've done a fragment talk although I…
@QuantumTessera @radscientist_ Gutted to have missed it! I think it's over 10 years since I've done a fragment talk although I have taken a few gratuitous pot shots at Ro3 and Fsp3 since then (most recently in Nature of LE): https://t.co/2xu214v5cB
@maria_denk Great Expectations (wished that Magwitch had strangled Pip in the graveyard). Catch 22 took me three attempts (the first of which was after grad school) but I introduced 'Nature of LE' with an adaptation of the prayer for "nice tight bombing pa
When claiming that an #ML model for binding affinity is explainable you need to aware that logarithmically-defined binding affinity can be set to zero by choosing the unit in which k.D is expressed(see NoLE) #cheminformatics #biophys #MedChem https://t.co
@biogerontology @FlyOnNo10Wall I included "the most interesting SAR is likely to be associated with the most deviant values" in Nature of Ligand Efficiency to celebrate England's ruling classes: https://t.co/2xu214v5cB
RT @pwk2013: @janhjensen Hopefully the authors are aware that the contribution of a protein-ligand contact to binding affinity is not an ex…
@janhjensen Hopefully the authors are aware that the contribution of a protein-ligand contact to binding affinity is not an experimental observable. 'Thermodynamic aspects of ligand–protein association' section in this article may be relevant: https://t.co
RT @pwk2013: @MedChemica The discussion of this study in 'Nature of Ligand Efficiency' may be of interest: https://t.co/2xu214v5cB
@MedChemica The discussion of this study in 'Nature of Ligand Efficiency' may be of interest: https://t.co/2xu214v5cB
@AmySYQin @iwatobipen @ChemRxiv There is a difference between predicting ΔG and capturing molecular interactions. Contribution of an intermolecular contact to ΔG is not generally an experimental observable as discussed in "Thermodynamic aspects of ligand–p
@TheDuckMachine To be punished by not being whipped is indeed bizarre and I included "the most interesting SAR is likely to be associated with the most deviant values" in Nature of Ligand Efficiency as a homage to the deviance of England's ruling class: ht
@med_chemist @KRHornberger @PubPeer Some of what is 'known' in #MedChem is actually based on flaky data analysis. My top gripe is #LigandEfficiency (perception of efficiency depends on the unit in which efficiency is expressed and this is physical nonsense
@SaraDoesScience Many medicinal chemists (and even cheminformaticians who really should know better) use a design metric (ligand efficiency) that causes perception to change when concentrations are expressed in different units: https://t.co/2xu214v5cB
RT @pwk2013: "While a structure–activity relationship (#SAR) can point to the importance of individual interactions, the contribution of a…
RT @pwk2013: "While a structure–activity relationship (#SAR) can point to the importance of individual interactions, the contribution of a…
"While a structure–activity relationship (#SAR) can point to the importance of individual interactions, the contribution of a protein–ligand contact to affinity is not, in general, an experimental observable" (re-post) #PhysChem #MedChem #cheminformatics h
@HewittPete @covid_moonshot The most interesting compounds beat the trends to the greatest extent (activity cliffs). Weak trends can still be used to normalize affinity as discussed in 'Alternatives to ligand efficiency for normalization of affinity' in ht
@med_chemist One of the problems with ligand efficiency (as conventionally defined) is that perception of efficiency varies with the unit used to define affinity. See Table 1 in https://t.co/2xu214v5cB
RT @pwk2013: Tolerance for bullshit (e.g. change in unit leading to change in perception) is an indication of weakness in a scientific disc…
Tolerance for bullshit (e.g. change in unit leading to change in perception) is an indication of weakness in a scientific discipline (re-post) #LigandEfficiency #MedChem #DrugDesign #DrugDiscovery #cheminformatics #chemoinformatics https://t.co/2xu214v5cB
@daniel_reker @friedo91 @rguha @macinchem Outliers can be seen as a type of 'activity cliff' and you can use a weak trend to normalize data. This is philosophy behind 'Andrews binding energy' and plotting potency against molecular size to observe residual
@mikeprime9 Can't remember if I ever mentioned that "... the most interesting SAR is likely to be associated with the most deviant values" was deliberately included in Nature of Ligand Efficiency to celebrate this sort of thing https://t.co/2xu214v5cB
@HeatherCarlsonh The initial slope in 1999 Kuntz et al study is influenced by an arbitrary choice of zero molecular size limit for affinity (this also appears to be the case for your 2012 study). This article (which proved too spicy for J Med Chem reviewer
@D_B_McConnell @BShoichet It may be a gem but the initial slope appears to be strongly influenced by what is an arbitrary choice of intercept. I'll mention @MichaelKGilson and there's 'Maximal affinity of ligands and fit quality' section in https://t.co/2
@pwk2013 Now I have time to read the whole of your article I am definitely going to use this in our new modelling and machine learning module! https://t.co/JBD8lrQmmt
Do you really want to base #MedChem decisions on a #metric that changes your perception when you express a quantity in a different unit? (re-post) #LigandEfficiency #DrugDesign #DrugDiscovery #cheminformatics #NotEvenWrong https://t.co/2xu214v5cB
@raelawrence1 @QuantumTessera I blame the extra-curricular activities of England's Ruling Classes, to which I paid homage in NoLE: "... the most interesting SAR is likely to be associated with the most deviant values" https://t.co/2xu214v5cB
@ChemBellenie @D_B_McConnell MW and other measures of molecular size can be useful to normalize activity (pic 1) and compare projects (pic 2) in fragment to lead work (for which MW usually increases significantly). Pics from Nature of #LigandEfficiency htt
@PM_orifice Knowing that knotted towels saved England from Boney & those other beastly foreigners prompted the homage "... the most interesting SAR is likely to be associated with the most deviant values" to England's Public Schools in The Nature of Li
@KRHornberger @D_B_McConnell Much of the bRo5 analysis is distinctly shaky and/or largely irrelevant to specific project #MedChem. See comment on attempts to set #LigandEfficiency threshold starting at: "The minimum Δg value of 0.12 kcal/mol per non-hydro
@ChemBellenie @Dereklowe Project teams often only measure solubility for most potent compounds. I'd suggest getting measurements for evenly spaced logP/D across range. Then fit logS to logP/D and examine residuals (see 'Alternatives to ligand efficiency..'
RT @pwk2013: Is sensitivity to criticism of #LigandEfficiency due to #metric being used to sell #FBDD to #DrugDiscovery leaders & investors…
Is sensitivity to criticism of #LigandEfficiency due to #metric being used to sell #FBDD to #DrugDiscovery leaders & investors? Low molecular complexity is stronger rationale for screening #fragments (re-post) #FBLD #MedChem #DrugDesign #cheminformatic
RT @pwk2013: Early morning at Berwick-on-Sea, North Coast Road, #Blanchisseuse, St George, #TrinidadAndTobago which is also the corresponde…
Early morning at Berwick-on-Sea, North Coast Road, #Blanchisseuse, St George, #TrinidadAndTobago which is also the correspondence address for the heretical Nature of #LigandEfficiency https://t.co/2xu214v5cB https://t.co/sm6zZxJR9d
RT @pwk2013: "In thermodynamic analysis, a change in perception resulting from a change in a standard state definition would generally be r…
RT @pwk2013: "In thermodynamic analysis, a change in perception resulting from a change in a standard state definition would generally be r…
"In thermodynamic analysis, a change in perception resulting from a change in a standard state definition would generally be regarded as a serious error rather than a penetrating insight." #DrugDesign #MedChem #cheminformatics https://t.co/2xu214v5cB
@dr_greg_landrum @ACSBioMed @ChEMBL I think you need to get out more 😀😀😀 The fundamental problem with LE is that it's meaningless because perception of efficiency changes when you use a different concentration unit to express affinity: https://t.co/2xu214
@souyakuchan @FrankvonDelft @jchodera @covid_moonshot @PostEra_AI @london_lab @griffen_ed Pharmacokinetic design can be seen as control of exposure and this is easier to achieve when the drug is highly potent. This is discussed in more detail in the introd
RT @pwk2013: The contribution of a contact between two regions of molecular surface to affinity (or stability) is not, in general, an exper…
The contribution of a contact between two regions of molecular surface to affinity (or stability) is not, in general, an experimental observable #PhysChem #BioPhys #CompChem #cheminformatics #MedChem #DrugDesign https://t.co/2xu214v5cB
@devivo_marco A wise user of ML models will always be concerned about applicability domain (e.g. does model work for my project compounds?) See "Sample bias can be significant, even in large datasets..." in https://t.co/2xu214dtO1
@ZondloLab @curiouswavefn I don't believe that ΔS can be distributed over non-covalent contacts in a unique manner. For protein-ligand binding, stoichiometry adds layer of complexity to problem. See 'Thermodynamic aspects of ligand–protein association' sec
@belledejour_uk I slipped a short tribute to England's ruling classes into my ligand efficiency article "... the most interesting SAR is likely to be associated with the most deviant values" https://t.co/2xu214v5cB
RT @pwk2013: "In #thermodynamic analysis, a change in perception resulting from a change in a standard state definition would generally be…
RT @pwk2013: "In #thermodynamic analysis, a change in perception resulting from a change in a standard state definition would generally be…
"In #thermodynamic analysis, a change in perception resulting from a change in a standard state definition would generally be regarded as a serious error rather than a penetrating insight" #DrugDesign #MedChem #cheminformatics #LigandEfficiency #LE https
@rflameiro @DennisWhom I think better to use logP than logD to normalize potency by lipophilicity (logD can be reduced simply by increasing extent of ionization and that's a design strategy which is likely to end in tears). Comments on LipE/LLE in this art
@GastreichM @D_B_McConnell The devil is in the details of the increments are defined. In thermodyamics, it usually ends in tears when one tries to decompose ∆G° into a sum of substructural terms (see critique of ref 54 in 'The nature of ligand efficiency')
@macinchem @D_B_McConnell The dependence of ∆G° on the standard concentration and the non-local nature (with respect to molecular interactions) of ∆S° mean that ∆G° cannot be decomposed into interaction-based contributions in a unique manner https://t.co/
@london_lab @UCSF @BShoichet @WeizmannScience @Dereklowe If you're planning to have a property-based design component (and you're also prepared to question some fundamental 'truths' of #DrugDiscovery) in your course at Weizmann Institute then this article
@rguha @wpwalters @JCIM_ACS @egonwillighagen @10705013 @BZdrazil If one can demonstrate relevance/utility with a non-proprietary datasets then why would one need to use proprietary datasets? Sample bias can be an issue even for large datasets as noted in
RT @pwk2013: "... readers might consider their likely responses to a hypothetical report that the space group for a crystal structure diffe…
"... readers might consider their likely responses to a hypothetical report that the space group for a crystal structure differed according to whether unit cell parameters were expressed in Ångstrom or in nanometer units" #cheminformatics #metric #LE http
That medicinal chemists use a design #metric that alters their perception when a different unit is used to express affinity points to a serious malaise in the #MedChem discipline #LigandEfficiency #cheminformatics #DrugDesign #DrugDiscovery https://t.co/2x
@BShoichet @london_lab @covid_moonshot @griffen_ed @PostEra_AI @OleinikovasV @PracticalFrag On the subject of LE, The Inhibitor of Promiscuity might wish to examine Table 1 in this article (which proved a little too spicy for J Med Chem) https://t.co/2xu2
@JCIM_ACS As conventionally defined, LE has a nontrivial dependence on the concentration used to define the standard state (or in which potency is expressed) and is physically meaningless https://t.co/2xu214v5cB
@RomanValiulin @ChemDraw I'm not sure that PSA or NRotB thresholds are included in Astex #Ro3 and you may also find comments made about Ro3 in this article to be helpful: https://t.co/2xu214dtO1
"Your job is to lead us in prayers, and from now on you’re going to lead us in a prayer for more #LigandEfficiency in every project. Is that clear? I think more ligand efficiency is something really worth praying for” (re-post) #DrugDesign #cheminformatics
@rachaelskyner @covid_moonshot Not valid to compare IC50 for reversible & irreversible inhibs. Would be prudent to check reversibility for quinolones. Problem with using LE as a design metric is that perception of efficiency changes with concentration
@MichelleFrancl @stuartcantrill The #PedantInChief may be interested to know that molarity (but not molality) varies with P and T and that perception of #LigandEfficiency varies with the unit in which affinity is expressed https://t.co/2xu214v5cB
@AverageEnsemble @web_tric @CBCErhardt @olexandr @JMedChem When preparing the preprint, I re-introduced (and 'enhanced') material that had really got the two reviewers spitting feathers and used appropriate countermeasures when I submitted to J Cheminf to
@Seigonie Greatly enjoyed seeing video in your weathercast and many thanks for featuring it. On an unrelated note, I've used my Blanchisseuse address for correspondence in two scientific publications, including this one (see author info near end of article