455
13 Continuous Molecular Fields Approach Applied to Structure-Activity Modeling
with one-class classification kernel-based methods for conducting virtual screening based on the similarity of molecular fields. The feasibility of this approach has
already been proved by us, see preliminary communication [13]. The rest of the
table shows the promising directions for further development of the CMF approach.
13.5 Conclusion
The CMF approach describes molecules by ensemble of continuous functions (molecular fields), instead of finite sets of molecular descriptors (such as interaction
energies computed at grid nodes). The potential advantages of this approach results
from the ability to approximate electronic molecular structures with any desirable
accuracy level, the ability to leverage the valuable information contained in partial derivatives of molecular fields (otherwise lost upon discretization) to analyze
models and enhance their predictive performance, the ability to apply integral transforms to molecular fields and models, etc.
The most attractive features of the CMF approach are its versatility and universality. By combining different types of molecular fields and methods of their approximation, different types of kernels with different types of kernel-based machine
learning methods, it is possible to present lots of existing methods in chemoinformatics and medicinal chemistry as particular cases within a universal methodology.
The CMF methodology can easily be extended to building classification and novelty detection models, visualizing them, performing virtual screening, processing
diverse datasets.
We see the following main directions for further development of the CMF approach: introduction of additional types of molecular fields, including conceptual
DFT molecular fields; tackling the issue of molecular alignment and flexibility; taking into account molecular flexibility; prediction of physicо-chemical properties of
chemical compounds; taking into account different ionization states of molecules,
their tautomers and conformers; extension of the approach to work with biological
macromolecules and supramolecular complexes; integration of special approaches
for preventing model selection bias; combining with different machine learning
methods aimed at solving various tasks.
The CMF approach is implemented as a set of scripts operating under the R environment for statistical computing and graphics [80]. A version of the software is
available at http://sites.google.com/site/conmolfields/.
Acknowledgments The authors thank Prof. Yu.A.Ustynyuk for stimulating discussion and
advice. The authors also thank Prof. A.Varnek and Dr. G.Marcou for valuable comments regarding the developed approach. This work was supported by Russian Foundation for Basic Research
(Grant 13-07-00511).
13 Continuous Molecular Fields Approach Applied to Structure-Activity Modeling
with one-class classification kernel-based methods for conducting virtual screening based on the similarity of molecular fields. The feasibility of this approach has
already been proved by us, see preliminary communication [13]. The rest of the
table shows the promising directions for further development of the CMF approach.
13.5 Conclusion
The CMF approach describes molecules by ensemble of continuous functions (molecular fields), instead of finite sets of molecular descriptors (such as interaction
energies computed at grid nodes). The potential advantages of this approach results
from the ability to approximate electronic molecular structures with any desirable
accuracy level, the ability to leverage the valuable information contained in partial derivatives of molecular fields (otherwise lost upon discretization) to analyze
models and enhance their predictive performance, the ability to apply integral transforms to molecular fields and models, etc.
The most attractive features of the CMF approach are its versatility and universality. By combining different types of molecular fields and methods of their approximation, different types of kernels with different types of kernel-based machine
learning methods, it is possible to present lots of existing methods in chemoinformatics and medicinal chemistry as particular cases within a universal methodology.
The CMF methodology can easily be extended to building classification and novelty detection models, visualizing them, performing virtual screening, processing
diverse datasets.
We see the following main directions for further development of the CMF approach: introduction of additional types of molecular fields, including conceptual
DFT molecular fields; tackling the issue of molecular alignment and flexibility; taking into account molecular flexibility; prediction of physicо-chemical properties of
chemical compounds; taking into account different ionization states of molecules,
their tautomers and conformers; extension of the approach to work with biological
macromolecules and supramolecular complexes; integration of special approaches
for preventing model selection bias; combining with different machine learning
methods aimed at solving various tasks.
The CMF approach is implemented as a set of scripts operating under the R environment for statistical computing and graphics [80]. A version of the software is
available at http://sites.google.com/site/conmolfields/.
Acknowledgments The authors thank Prof. Yu.A.Ustynyuk for stimulating discussion and
advice. The authors also thank Prof. A.Varnek and Dr. G.Marcou for valuable comments regarding the developed approach. This work was supported by Russian Foundation for Basic Research
(Grant 13-07-00511).