Proposal to built an interface for SADAPT

Dear colleagues POP-PHARM would like to know if there would be interest from the Pharmacometric community in the potential development of a user-friendly interface that would facilitate the model building in SADAPT. The template interface would be the program I developed with its interface (Iconus) (PDX-MC-PEM) and automatic scripting that would built a model template which would include for examples automatic code for population mixture, interoccasion and covariate analysis (assuming log-linear regression for example). Simple questions would be answered by the user and he would have the option to either run the analysis with my program or be connected to the SADAPT engine that is usually more powerful and gives all the outputs that are usually needed (like standard errors, post-hoc etc...) This project can become reality if the Population community is interested and would be available free of charge. This will become reality once I am getting enough interest from the community in building this kind of facilitator. I would appreciate if anybody interested could give me some feedback with also desire features they would like to have implemented. As you know, we have planned a SADAPT workshop on April 14-15 ( http://www.mosaicnj.org/2009/02/21/s-adapt-intermediate-workshop ) and anybody coming to the workshop could ask me at that time any questions about this new potential project. If you need more information about the upcoming SADAPT workshop, please visit http://www.mosaicnj.org/2009/02/21/s-adapt-intermediate-workshop to access to both the agenda and registration. Best Regards; Serge Guzy; Ph.D President, CEO; POP-PHARM; Inc; cel (510 542 76 48) -- The information contained in this email message may contain confidential or legally privileged information and is intended solely for the use of the named recipient(s). No confidentiality or privilege is waived or lost by any transmission error. If the reader of this message is not the intended recipient, please immediately delete the e-mail and all copies of it from your system, destroy any hard copies of it and notify the sender either by telephone or return e-mail. Any direct or indirect use, disclosure, distribution, printing, or copying of any part of this message is prohibited. Any views expressed in this message are those of the individual sender, except where the message states otherwise and the sender is authorized to state them to be the views of XOMA.

Serge, Thank you for your offer to put time and energy into making S-ADAPT more usable. I have had limited experience with S-ADAPT but the nightmarish complexity of having to write multiple subroutines in different files and special data sets to describe covariates made me give up. I was assisted by someone with quite a lot of S-ADAPT experience but after some months of effort we could not get what seemed to me to be a simple model to work. The same model and covariate structure was easily expressed using NM-TRAN for NONMEM. I was epecially interested in using S-ADAPT because of its support for parallel thread execution on multiple CPU/core systems. Run times for complex problems are my biggest difficulty at the moment. Among mixed effect modelling programs only S-ADAPT supports parallel execution and as far as I know the other main programs are still stuck with a single threaded programming model for the forseeable future. My suggestion would be to work on creating a scripting language similar to NM-TRAN that allows the user to specify the data, the parameters and the model in a SINGLE text file script. In principle this text based script would allow a smart programmer like yourself to write all the subroutines and create the special data sets just like NM-TRAN does for NONMEM users. If S-ADAPT had a coherent and integrated text based script then this would make it far more usable to people like me who cannot afford to spend weeks developing the code and data for just the base model. If necesssary developers could write GUI front ends to hold the hands of beginners who are not comfortable with writing the full script. In the long run it should be possible to create a common mixed effect modelling language that would be independent of the backend estimation program. MLX-TRAN for Monolix has made some progress in that direction. The NLME consortium ( http://www.nlme.org/ ) has proposed in some detail the steps needed to do this and there is strong interest in persuing this idea. Best wishes, Nick Serge Guzy wrote: > Dear colleagues > > POP-PHARM would like to know if there would be interest from the Pharmacometric community in the potential development of a user-friendly interface that would facilitate the model building in SADAPT. > > The template interface would be the program I developed with its interface (Iconus) (PDX-MC-PEM) and automatic scripting that would built a model template which would include for examples automatic code for population mixture, interoccasion and covariate analysis (assuming log-linear regression for example). Simple questions would be answered by the user and he would have the option to either run the analysis with my program or be connected to the SADAPT engine that is usually more powerful and gives all the outputs that are usually needed (like standard errors, post-hoc etc…) > > This project can become reality if the Population community is interested and would be available free of charge. > > This will become reality once I am getting enough interest from the community in building this kind of facilitator. > > I would appreciate if anybody interested could give me some feedback with also desire features they would like to have implemented. > > As you know, we have planned a SADAPT workshop on April 14-15 ( http://www.mosaicnj.org/2009/02/21/s-adapt-intermediate-workshop ) and anybody coming to the workshop could ask me at that time any questions about this new potential project. If you need more information about the upcoming SADAPT workshop, please visit http://www.mosaicnj.org/2009/02/21/s-adapt-intermediate-workshop to access to both the agenda and registration. > > Best Regards; > > Serge Guzy; Ph.D > > President, CEO; POP-PHARM; Inc; > > cel (510 542 76 48) > > ------------------------------------------------------------------------ > > The information contained in this email message may contain confidential or legally privileged information and is intended solely for the use of the named recipient(s). No confidentiality or privilege is waived or lost by any transmission error. If the reader of this message is not the intended recipient, please immediately delete the e-mail and all copies of it from your system, destroy any hard copies of it and notify the sender either by telephone or return e-mail. Any direct or indirect use, disclosure, distribution, printing, or copying of any part of this message is prohibited. Any views expressed in this message are those of the individual sender, except where the message states otherwise and the sender is authorized to state them to be the views of XOMA. -- Nick Holford, Dept Pharmacology & Clinical Pharmacology University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New Zealand [email protected] tel:+64(9)923-6730 fax:+64(9)373-7090 http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford

Nick, S-ADAPT is not the only parallel NLME system currently available or on the near term horizon. Pharsight currently has a pharmacometric NLME system ("Phoenix") under late stage development that is multithreaded (see PAGE abstract 979, Leary and Dunlavey, "Parallel Processing for Nonlinear MIxed effects Modeling", 2006) The parallelization is over the subjects, so the computations required to find the log likelihood contribution of each subject during an evaluation of the population log likelihood can be performed on separate threads (and hence separate processors). The threads communicate via the MPI (message passing interface) library (the de facto standard for distributed memory parallelization as for example, in a cluster, although it works equally well for shared memory systems such as current generation dual and quad processor chips. ) Under the right cirucumstances (large numbers of subjects Nsub >>Nprocessors, with reasonably balanced data per subject ), the parallel speedup can be a fairly large fraction of Nprocessors. It has both a GUI for model building and expression, as well as a text based modeling language that is similar in expressiveness and functionality to NMTRAN, although quite a bit different in syntax. Also, the NPEM and NPAG nonparametric software from Roger Jelliffe's LAPK lab has long been parallelized vis MPI and successfully run on a huge variety of parallel shared and distributed memory systems. Bob Leary

Dear Nick Thanks for your useful comments and proposals. I think that there a few steps to be done sequentially in order to reach your goal. Obviously I do not have the skills to create this kind of translator but I wanted first to minimize the need to write all these SADAPT subroutines but only a few that match more with what you are doing when using NONMEM. Since my intention is to provide this kind of facilitator free of charge, it is not easy for me to find support in going beyond the basic idea and create what you suggested. If I can get support for that project, then I can help a lot in making that happening. About you not succeeding with a basic problem with SADAPT, it is too bad you did not ask for my help. I am 100% sure I can fix the code and getting the results you would expect with SADAPT. If you have still this problem ready to be tried (dataset and model file), then I would be happy to take a look and sending you the fixed code with the results. I think that SADAPT offers more than just parallel processing capability. SADAPT has the advantage to work interactively while NONMEM itself still stays more a kind of batch program. You start your problem, you look at the results and stop whenever you want, make the changes based on what you saw and then continue. You have the option to learn during your exploratory analysis and finally getting optimal results. Another advantage is the use of scripting code that are an integral part of SADAPT and allow you to run in a batch mode many different problems one after the other. For example, you can perform a covariate analysis starting with the full model and then write a simple script that will emulate the backward covariate deletion method. All the outputs are stored automatically and you just have to compare all the log-likelihood to decide who the remaining covariate parameters are. My understanding about NONMEM 7 is that NONMEM 7 will not have the interactive capability like SADAPT, neither the scripting advantage SADAPT has. Many new methods we (Bob Bauer and I) implemented in the last 8 years will be available but not scripting and interactive capability. I think also that the SADAPT offers an integrator engine that is more powerful than the NONMEM one. What is missing with SADAPT is a way to help the user writing the model file and that is what I wanted to concentrate on. The back end of SADAPT is I think very good as all the outputs files are automatically created from which you could use other programs if necessary to have optimal diagnostic tools. Anybody that want help with first time SADAPT problem solving, I will be happy to assist free of charge and making their first problem working. In addition, we provide many workshops at a very cheap price where all the features of SADAPT are taught. Step by step examples from basic to advanced problems can be grabbed from my site (just ask me) and you can learn either by your self or through workshops how to use optimally SADAPT. Best Regards; Serge Guzy; President, POP-PHARM; Inc;

Dear Nick Thanks for your useful comments and proposals. I think that there a few steps to be done sequentially in order to reach your goal. Obviously I do not have the skills to create this kind of translator but I wanted first to minimize the need to write all these SADAPT subroutines but only a few that match more with what you are doing when using NONMEM. Since my intention is to provide this kind of facilitator free of charge, it is not easy for me to find support in going beyond the basic idea and create what you suggested. If I can get support for that project, then I can help a lot in making that happening. About you not succeeding with a basic problem with SADAPT, it is too bad you did not ask for my help. I am 100% sure I can fix the code and getting the results you would expect with SADAPT. If you have still this problem ready to be tried (dataset and model file), then I would be happy to take a look and sending you the fixed code with the results. I think that SADAPT offers more than just parallel processing capability. SADAPT has the advantage to work interactively while NONMEM itself still stays more a kind of batch program. You start your problem, you look at the results and stop whenever you want, make the changes based on what you saw and then continue. You have the option to learn during your exploratory analysis and finally getting optimal results. Another advantage is the use of scripting code that are an integral part of SADAPT and allow you to run in a batch mode many different problems one after the other. For example, you can perform a covariate analysis starting with the full model and then write a simple script that will emulate the backward covariate deletion method. All the outputs are stored automatically and you just have to compare all the log-likelihood to decide who the remaining covariate parameters are. My understanding about NONMEM 7 is that NONMEM 7 will not have the interactive capability like SADAPT, neither the scripting advantage SADAPT has. Many new methods we (Bob Bauer and I) implemented in the last 8 years will be available but not scripting and interactive capability. I think also that the SADAPT offers an integrator engine that is more powerful than the NONMEM one. What is missing with SADAPT is a way to help the user writing the model file and that is what I wanted to concentrate on. The back end of SADAPT is I think very good as all the outputs files are automatically created from which you could use other programs if necessary to have optimal diagnostic tools. Anybody that want help with first time SADAPT problem solving, I will be happy to assist free of charge and making their first problem working. In addition, we provide many workshops at a very cheap price where all the features of SADAPT are taught. Step by step examples from basic to advanced problems can be grabbed from my site (just ask me) and you can learn either by your self or through workshops how to use optimally SADAPT. Best Regards; Serge Guzy; President, POP-PHARM; Inc;

Bob, Thanks for letting me know that NPEM/NPAG are parallelised. I know you told me about that a meeting some years ago and I am afraid I had forgotten it. I guess I just had'nt heard of anybody doing it to remind me. My ignorance I am afraid. As for Phoenix -- I look forward to seeing it and using it. But at present it remains off the horizon... Nick Bob Leary wrote: > Nick, > > S-ADAPT is not the only parallel NLME system currently available or on the near term horizon. Pharsight currently has a pharmacometric NLME system ("Phoenix") under late stage development that is multithreaded (see PAGE abstract 979, Leary and Dunlavey, "Parallel Processing for Nonlinear MIxed effects Modeling", 2006) The parallelization is over the subjects, so the computations required to find the log likelihood contribution of each subject during an evaluation of the population log likelihood can be performed on separate threads (and hence separate processors). The threads communicate via the MPI (message passing interface) library (the de facto standard for distributed memory parallelization as for example, in a cluster, although it works equally well for shared memory systems such as current generation dual and quad processor chips. ) Under the right cirucumstances (large numbers of subjects Nsub >>Nprocessors, with reasonably balanced data per subject ), the parallel speedup can be a fairly large fraction of Nprocessors. It has both a GUI for model building and expression, as well as a text based modeling language that is similar in expressiveness and functionality to NMTRAN, although quite a bit different in syntax. Also, the NPEM and NPAG nonparametric software from Roger Jelliffe's LAPK lab has long been parallelized vis MPI and successfully run on a huge variety of parallel shared and distributed memory systems. > > Bob Leary > ------------------------------------------------------------------------ > *From:* [email protected] on behalf of Nick Holford > *Sent:* Sun 3/22/2009 4:04 AM > *To:* [email protected] > *Subject:* Re: [NMusers] Proposal to built an interface for SADAPT > > Serge, > > Thank you for your offer to put time and energy into making S-ADAPT more > usable. I have had limited experience with S-ADAPT but the nightmarish > complexity of having to write multiple subroutines in different files > and special data sets to describe covariates made me give up. I was > assisted by someone with quite a lot of S-ADAPT experience but after > some months of effort we could not get what seemed to me to be a simple > model to work. The same model and covariate structure was easily > expressed using NM-TRAN for NONMEM. > > I was epecially interested in using S-ADAPT because of its support for > parallel thread execution on multiple CPU/core systems. Run times for > complex problems are my biggest difficulty at the moment. Among mixed > effect modelling programs only S-ADAPT supports parallel execution and > as far as I know the other main programs are still stuck with a single > threaded programming model for the forseeable future. > > My suggestion would be to work on creating a scripting language similar > to NM-TRAN that allows the user to specify the data, the parameters and > the model in a SINGLE text file script. In principle this text based > script would allow a smart programmer like yourself to write all the > subroutines and create the special data sets just like NM-TRAN does for > NONMEM users. > > If S-ADAPT had a coherent and integrated text based script then this > would make it far more usable to people like me who cannot afford to > spend weeks developing the code and data for just the base model. If > necesssary developers could write GUI front ends to hold the hands of > beginners who are not comfortable with writing the full script. > > In the long run it should be possible to create a common mixed effect > modelling language that would be independent of the backend estimation > program. MLX-TRAN for Monolix has made some progress in that direction. > The NLME consortium ( http://www.nlme.org/) has proposed in some detail > the steps needed to do this and there is strong interest in persuing > this idea. > > Best wishes, > > Nick > > Serge Guzy wrote: > > > > Dear colleagues > > > > POP-PHARM would like to know if there would be interest from the > > Pharmacometric community in the potential development of a > > user-friendly interface that would facilitate the model building in > > SADAPT. > > > > The template interface would be the program I developed with its > > interface (Iconus) (PDX-MC-PEM) and automatic scripting that would > > built a model template which would include for examples automatic code > > for population mixture, interoccasion and covariate analysis (assuming > > log-linear regression for example). Simple questions would be answered > > by the user and he would have the option to either run the analysis > > with my program or be connected to the SADAPT engine that is usually > > more powerful and gives all the outputs that are usually needed (like > > standard errors, post-hoc etc…) > > > > This project can become reality if the Population community is > > interested and would be available free of charge. > > > > This will become reality once I am getting enough interest from the > > community in building this kind of facilitator. > > > > I would appreciate if anybody interested could give me some feedback > > with also desire features they would like to have implemented. > > > > As you know, we have planned a SADAPT workshop on April 14-15 ( > > http://www.mosaicnj.org/2009/02/21/s-adapt-intermediate-workshop ) and > > anybody coming to the workshop could ask me at that time any questions > > about this new potential project. If you need more information about > > the upcoming SADAPT workshop, please visit > > http://www.mosaicnj.org/2009/02/21/s-adapt-intermediate-workshop to > > access to both the agenda and registration. > > > > Best Regards; > > > > Serge Guzy; Ph.D > > > > President, CEO; POP-PHARM; Inc; > > > > cel (510 542 76 48) > > > > > > ------------------------------------------------------------------------ > > The information contained in this email message may contain > > confidential or legally privileged information and is intended solely > > for the use of the named recipient(s). No confidentiality or privilege > > is waived or lost by any transmission error. If the reader of this > > message is not the intended recipient, please immediately delete the > > e-mail and all copies of it from your system, destroy any hard copies > > of it and notify the sender either by telephone or return e-mail. Any > > direct or indirect use, disclosure, distribution, printing, or copying > > of any part of this message is prohibited. Any views expressed in this > > message are those of the individual sender, except where the message > > states otherwise and the sender is authorized to state them to be the > > views of XOMA. > > -- > Nick Holford, Dept Pharmacology & Clinical Pharmacology > > University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New Zealand > > [email protected] tel:+64(9)923-6730 fax:+64(9)373-7090 > http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford -- Nick Holford, Dept Pharmacology & Clinical Pharmacology University of Auckland, 85 Park Rd, Private Bag 92019, Auckland, New Zealand [email protected] tel:+64(9)923-6730 fax:+64(9)373-7090 http://www.fmhs.auckland.ac.nz/sms/pharmacology/holford