Visualization of Multi-Property Landscapes for Compound Selection and Optimization

Antonio de la Vega de Leon, Shilva Kayastha, Dilyana Dimova, Thomas Schultz und Jürgen Bajorath
In: J. Computer Aided Molecular Design (2015)
 

Abstract

Compound optimization generally requires considering multiple properties in concert and reaching a balance between them. Computationally, this process can be supported by multi-objective optimization methods that produce numerical solutions to an optimization task. Since a variety of comparable multi-property solutions are usually obtained further prioritization is required. However, the underlying multi-dimensional property spaces are typically complex and difficult to rationalize. Herein, an approach is introduced to visualize multi-property landscapes by adapting the concepts of star and parallel coordinates from computer graphics. The visualization method is designed to complement multi-objective compound optimization. We show that visualization makes it possible to further distinguish between numerically equivalent optimization solutions and helps to select drug-like compounds from multi-dimensional property spaces. The visualization methodology is intuitive, applicable to a wide range of chemical optimization problems, and made freely available to the scientific community.

Stichwörter: activity landscapes, Compound optimization, multi-objective optimization, multi-property landscapes, structure-property relationships, Visualization

Bilder

Bibtex

@ARTICLE{delaVegadeLeon2015JCAMD,
    author = {de la Vega de Leon, Antonio and Kayastha, Shilva and Dimova, Dilyana and Schultz, Thomas and
              Bajorath, J{\"u}rgen},
     title = {Visualization of Multi-Property Landscapes for Compound Selection and Optimization},
   journal = {J. Computer Aided Molecular Design},
      year = {2015},
      note = {To appear.},
  keywords = {activity landscapes, Compound optimization, multi-objective optimization, multi-property landscapes,
              structure-property relationships, Visualization},
  abstract = {Compound optimization generally requires considering multiple properties in concert and reaching a
              balance between them. Computationally, this process can be supported by multi-objective optimization
              methods that produce numerical solutions to an optimization task. Since a variety of comparable
              multi-property solutions are usually obtained further prioritization is required. However, the
              underlying multi-dimensional property spaces are typically complex and difficult to rationalize.
              Herein, an approach is introduced to visualize multi-property landscapes by adapting the concepts of
              star and parallel coordinates from computer graphics. The visualization method is designed to
              complement multi-objective compound optimization. We show that visualization makes it possible to
              further distinguish between numerically equivalent optimization solutions and helps to select
              drug-like compounds from multi-dimensional property spaces. The visualization methodology is
              intuitive, applicable to a wide range of chemical optimization problems, and made freely available
              to the scientific community.}
}