The ability to predict and quantify solvation effects like hydrogen bonding can be of great value in many areas of computational, medicinal, and physical chemistry. The conductor-like screening model for realistic solvation (COSMO-RS) combines a quantum chemical treatment of solutes and solvents with an efficient statistical thermodynamics procedure for the molecular surface interactions to enable the calculation of thermophysical data for liquid systems that other methods can barely predict.
Professor Andreas Klamt, creator of the COSMO-RS method, and CEO of COSMOlogic GmbH & Co. KG, Germany, tells Dr. Vera Köster of ChemViews Magazine how he came up with the idea behind COSMO-RS, how he transformed this into a thriving business, and discusses his views on where the field of computational chemistry is heading.
Can you please introduce your company, COSMOlogic, in a few words?
COSMOlogic is a still quite young and steadily growing software company, providing software for the prediction of fluid phase thermodynamics data, e.g., solubilities, partition and activity coefficients, vapor pressures, etc., based on quantum chemical calculation.
Where did the idea for COSMOlogic originate?
Usually quantum chemists know little about fluid phase thermodynamics, although this is of fundamental importance for almost all areas of chemistry. On the other hand, chemical engineers usually know a lot about fluid phase thermodynamics, but little about quantum chemistry (QC). Based on my sound fundament as theoretical physicist, I learned a bit about QC in my 12 years at Bayer, and there I happened to invent a very efficient variant of the dielectric continuum solvent models, which are quite popular in QC.
Later I realized the fundamental shortcomings of the dielectric continuum approach, including my COSMO model, and I developed a unique combination of this concept with fluid phase thermodynamics, named COSMO-RS (COnductorlike Screening MOdel for Realistic Solvation). And suddenly I had bridged the gap between QC and fluid phase thermodynamics.
So COSMO-RS was the starting point of your company?
Yes, definitely. In 1999 Professor Wolfgang Arlt, at that time at the Technical University Berlin, now at the University of Erlangen, both Germany, proved the applicability of COSMO-RS in chemical engineering. From that it was clear that this has a lot of additional industrial potential.
Is it still most important for your business?
COSMO-RS in form of the COSMOtherm program is still the core technology of COSMOlogic. We have built and are still extending a lot of additional functionality, application areas, and graphical user interfaces, but all is around COSMO-RS. Nevertheless, the distribution and customization of the quantum chamical program package TURBOMOLE, which originally was developed at University of Karlsruhe, Germany, and which is our work horse for the quantum part of our business, has become a nice side business.
What are the key competitive features of your software?
A sound physical basis, robustness, vividness, and applicability of the same basic principle to a wide range of properties, ranging from chemical engineering phase diagrams for separation at the one end to the binding of ligands into enzyme receptors at the other end.
How do the predicted properties compare to experimental data?
Our accuracy is roughly 0.4 kcal/mol for the free energy difference of a molecule between two phases, which results in about 0.3 log-units error. This means that we have up to a factor of 2 deviation for the corresponding equilibrium constant, e.g., for a partition coefficient or vapor pressure. This may sound a lot to many people, but it currently is the best available predictive method, as we have shown in several blind test predictions. The advantage of our COSMO-RS is that this error estimate is very robust with respect to a large number of properties and a really broad range of chemistry. And this is still useful for screening purposes. As I mentioned before, we cannot predict the last 10 % difference.
I recommend to be always skeptical if a method predicts logarithmic partition coefficients, solubilities or vapor pressures within better than 0.2 log-units. Such accuracy currently can only be achieved by massive training and parameterization on existing experimental data. But such methods often fail dramatically if they are faced with an unusual situation, because they have no physics in their equations. Unfortunately there are a few vendors, who are extremely good in training their flexible methods on almost all published experimental data. Any really predictive methods thus must fail in any comparison with such method, if the comparison is – as usual – done on published data sets. But if you confront such software with new in-house data, it often fails dramatically.
Is COSMO-RS also available online?
No, but there is a free demo-version available on request. And this demo version already allows for a wide range of nice applications and it can be used very nicely for teaching purposes.
Besides selling a software product, do you also sell a service?
Mostly we sell software, but we also sell contract calculations, as well as consulting, depending on the capabilities and needs of the customer.
If you do contract research, how can you guarantee confidentiality?
We have a good and reliable staff here with minimal fluctuations, and we have completely isolated our servers from the internet.
Do you have your own computer center?
Yes, a heterogeneous collection of LINUX boxes with multiple cores, a few Windows machines and Macs, in order to be able to develop and test our software on very different platforms. Altogether roughly 30 computer cores. This is not really a computer center, but the computational needs of our software are limited and we do not need more.
What kind of companies do you work for?
In addition to the large number of academic and public research institutions, we have about 80 companies worldwide using our software. These range from big chemical companies, such as BASF, Dow, Dupont, Evonik, …, through big pharma, e.g., Pfizer, BMS, AstraZeneca, … and consumer care companies like Procter&Gamble or Unilever, down to small research oriented companies such as Origenis in Munich, Germany.
How do these companies apply computational chemistry methods in their research?
Most of our industrial companies already have computational chemistry departments and use our COSMOtherm software as an additional tool. This is especially valuable for the calculation of all kinds of properties in liquid phases. But some companies even start to do computational chemistry with COSMOtherm and the underlying TURBOMOLE software.
To which fields of industry is computational chemistry most relevant?
The answer to this question will vary depending on whom you ask. Historically pharmaceutical chemistry has the largest experience in this area and definitely the largest number of computational chemists in industry will be in pharma. Nevertheless, based on my own experience at Bayer, and talking to so many colleagues in industry during the past 13 years, I know that there is a lot of potential in other areas of industrial chemistry and chemical engineering. BASF surely has the highest competence in this area.
Where do you see the field heading?
Quantum chemistry based computational chemistry will become a standard tool for prescreening in many areas of chemistry. Calculations with sufficient accuracy for screening purposes can be done nowadays routinely for large numbers of compounds. Hence questions like: “Which pure or mixed solvents might be good candidates for this or that purpose?” can be answered with moderate costs, and much faster than by experimental screening. Nevertheless, at the end these candidates will have to be tested and finally optimized in the lab, since computational chemistry is not yet accurate enough in order to optimize within the last 10 %.
What made you interested in founding your own company?
After being head of the central computational chemistry unit at Bayer for three years, Bayer – as many other companies – split the central research resources into small parts and associated them to the business units. I was not happy about that and still believe that a lot of valuable expertise, built over decades, got lost in that process. Anyway, looking for a model which would allow me to continue research in computational chemistry, I saw the options of taking an academic position or starting my own company. The academic option did not appear to be realistic on a short term, and hence I made a business plan evaluating the entrepreneur option. I was very confident that this should work based on my unique solvation modeling methods and the 12 years of industrial computational chemistry experience. And it works!
Meanwhile, I am glad that I have my own company, do not have to apply for grants and can do the research I am interested in based on the revenues of my company. The academic appointment as honorary professor in physical chemistry in Regensburg, Germany, came later based on an “after-seminar beer”. I like teaching one week a year. That is fun, but also enough.
What is the biggest hurdle you have faced or are still facing?
The right balance between publication and proprietary knowledge. On the one hand, we are doing fundamental research, and we really like to publish our findings, as I did from the beginning with my COSMO and COSMO-RS methods. On the other hand, I had to learn that some people in academia, even highly respected peers, and maybe even those from Germany, have no scruples to re-implement your methods in a Ph.D. thesis funded with public money, and later make the remake available either for free or via a commercial company. Fortunately, none of the remakes has been developed with similar rigor and continuity as COSMOtherm, and therefore they are not really harming us. But such events make you think about keeping some details unpublished for a while.
Why did you decide to become a computational chemist?
By education I am a theoretical solid state physicist. But after finishing my Ph.D. it was the German chemical industry that showed the largest interest in my skills. And thus I ended up in computational chemistry. I never regretted that. Computational chemistry is a fascinating area, with so many interesting applications and such fast development.
What fascinates you most about your job?
Being able to make a living from my own ideas, the freedom to follow new ideas as much as I like, and the opportunity to discuss so many interesting projects with so many different potential and existing customers from all areas of chemistry.
Thank you for the interview.
Andreas Klamt studied physics at the University of Göttingen, Germany, and gained his Ph.D. in theoretical solid state physics at the Max-Planck-Institute and the University of Stuttgart, Germany, in 1987. He joined Bayer AG, Leverkusen, Germany, in 1987 as a Computational Chemist and was promoted to the Head of the Central Computational Chemistry Department there in 1996. He left Bayer in 1999 to found COSMOlogic GmbH & Co. KG, Leverkusen, Germany, where he is currently CEO.
Klamt gained his Habilitation in 2005, resulting in his appointment as a private lecturer (adjunct professor) at the University of Regensburg, Germany, from 2005–2012, and his current position as honorary professor at the same university.
Selected Publications
- COSMO: A New Approach to Dielectric Screening in Solvents with Explicit Expressions for the Screening Energy and its Gradient,
A. Klamt, G. Schüürmann,
J. Chem. Soc. Perkin Trans. 2, 1993, 799–805.
DOI: 10.1039/p29930000799 - Conductor-like Screening Model for Real Solvents: A New Approach to the Quantitative Calculation of Solvation Phenomena,
A. Klamt,
J. Phys. Chem. 1995, 99, 2224–2235.
DOI: 10.1021/j100007a062 - COSMO-RS: An Alternative to Simulation for Calculating Thermodynamic Properties of Liquid Mixtures,
A. Klamt, F. Eckert, W. Arlt,
Annual Review of Chemical and Biochemical Engineering, 2010, 1,101–122.
DOI: 10.1146/annurev-chembioeng-073009-100903 - Polarization charge densities provide a predictive quantification of hydrogen bond energies,
A. Klamt, J. Reinisch, F. Eckert, A. Hellweg, M. Diedenhofen,
Phys. Chem. Chem. Phys. 2012, 14, 955–963.
DOI: 10.1039/C1CP22640A - COSMOsar3D: Molecular Field Analysis Based on Local COSMO σ-Profiles,
A. Klamt, M. Thormann , K. Wichmann, P. Tosco,
J. Chem. Inf. Model. 2012.
DOI: http://dx.doi.org/10.1021/ci300231t
Selected Awards
- Winner of the 1st Industrial Fluid Phase Properties Prediction Challenge (AICHE/NIST), Nov. 2002
- Winner of the 6th Industrial Fluid Phase Properties Prediction Challenge (AICHE/NIST), Nov. 2010
- 2012 EFCE Distinguished Lecture Award in Thermodynamics and Transport Properties, Oct. 2012
