From:Toufigh GordiSubject:[NMusers] Michaelis-Menten model i nonmem Date:Mon, 17 Feb 2003 11:25:39 -0500 Hi! I am trying to implement a "simple" binding action of a drug in nonmem. The PK of the drug is characterized (3-comp). The binding is given as %bound receptors (between 0-100). I have the following coding: $ERROR IPRED=F W=1 IRES=DV-IPRED IWRES=IRES/W Y1=IPRED+(IPRED*EPS(1)+EPS(2))*W Y2=(BMAX*IPRED/(BC50+IPRED))+EPS(3) Q=1 IF (PKPD.EQ.2) Q=0 Y=Q*Y1+(1-Q)*Y2 All PK parameters (including eps (1) and (2)) are fixed. Altough the model gives expected estimates of BMAX and BC50, EPS(3) is large (over 300%), which results in poor individual fits. Any suggestions? Toufigh Gordi _______________________________________________________ From: Nick Holford Subject: Re: [NMusers] Michaelis-Menten model i nonmem Date:Tue, 18 Feb 2003 08:08:23 +1300 Toufigh, First suggestion is to make your model account for reality more closely and put in a term for non-specific binding. I also wonder how you know the % bound (in order to compute the DV for bound conc) if you don't already know the binding parameters? It's often a good idea to fit the raw binding observation rather than transforming it into a % of something that is usually not really known very well. Also remember that the binding model assumes CONC is unbound. Is this really the case for the conc that your PK model is predicting? Or is this total (bound+unbound) conc? You might also consider a proportional residual error for the binding prediction. Finally, there is no need for the variables Q and W in your code. It can be expressed more clearly (IMHO) as follows: $ERROR CONC=F IF (PKPD.EQ.2) THEN Y=BMAX*CONC/(BC50+CONC) + CONC*NS + EPS(3) ; NS is non-specific binding parameter ELSE Y=CONC*(1+EPS(1)) + EPS(2) ENDIF _______________________________________________________