Developing data-driven models to identify environmentally friendly degradation modulators


- Tim Würger, M.Sc., Institute of Polymer Composites -

Magnesium (Mg) bears versatile material properties and great potential to become the material of the future as it can be used in a variety of applications. It can be employed as structural material in  the aerospace and automotive industries, as a base material for bioabsorbable medical implants and as an anode material for primary batteries (see Figure 1). Among the biggest challenges in this context is gaining control over corrosion properties to unlock the full potential of magnesium for a specific target application. For example, for transportation applications, degradation of the material must be prevented to avoid critical material defects. In this area in particular, environmentally friendly alternatives to the very efficient but highly toxic Cr(VI) corrosion protection strategies must be found, as their use will be banned in the near future due to a new REACH regulation. For medical applications, the degradation rate of the material must be modified to allow the injury to heal before the implant dissolves. Finally, for battery applications, continuous dissolution of the anode material must be ensured to keep the battery voltage constant. Consequently, benign degradation modulating strategies are required.

One promising approach is the use of small organic compounds as degradation modulators, which can be introduced either via a coating on the base material or as a component of the used electrolyte solution. However, the sheer number of organic compounds with potentially useful properties is far too large to identify promising candidates through an exclusively experimental approach. A possible solution to this challenge is to develop models based on quantitative structure-property relationships (QSPR) in order to explore the chemical space more quickly and sustainably. These models offer great opportunities to reduce the chemical space of potential candidates by preselecting suitable additives for experimental investigation.[1,2] In a comprehensive experimental study employing hydrogen evolution experiments, Lamaka et al. measured the corrosion inhibition performance of over 150 organic compounds for nine distinct Mg-based materials, thus providing a suitable training set for a QSPR model.[3]

For instance, molecular similarity measures can be combined with dimensionality reduction algorithms in order to create two-dimensional similarity maps, such as sketch-maps (see Figure 2) [4]. Here, each dot corresponds to a molecular structure. Two dots lying close or far away to each other indicate that their associated molecular structures are similar or dissimilar, respectively. After coloring the dots according to a target property of interest, potentially formed clusters can indicate a relationship between molecular structure and e.g., the corrosion inhibition performance.

The degradation modulating properties of yet untested compounds (unrimmed dots) can then be predicted in two ways: qualitatively or quantitatively. On the one hand, compounds not included in the training dataset can be mapped onto the sketch-map where the relative position to priorly identified clusters allows a qualitative prediction.[1] On the other hand, molecular similarites can be used as direct input for a kernel ridge regression (KRR) model to quantitatively predict the inhibition efficiency (colored, unrimmed dots).[5]

As the identification of potentially useful compounds can be tedious and time-consuming due to the immense size of the chemical space, methods for preselection of potential candidates are of great interest. By exploiting both of the previously described predictive methods, a workflow could be developed that provides automated selection of untested compounds with promising properties for experimental testing by screening a large commercial molecular database.[5]  The workflow is implemented in the web app "ExChem" which can be found at .

The close collaboration with the Institute of Surface Science at HZG enables the experimental validation of the effect on the degradation behavior of magnesium predicted by machine learning as well as the further development of the already established models.[2,5] These synergistic computational approaches should significantly improve the model interpretation and predictive power of the underlying machine learning models, thus paving the way for the discovery or rational design of tailored Mg dissolution agents.

More Information