Fluorescent molecular probes with emission in the near-infrared region represent
a promising diagnostic method for detecting amyloid b aggregates,
biomarkers characteristic of Alzheimer’s disease. Since both the preparation
and measurement of the optical properties of such molecules are timeconsuming
and costly, it would be highly efficient if one could predict the
optical properties of these compounds in contact with amyloid b aggregates
based on their molecular structure, thereby reducing the number of unnecessary
syntheses and characterizations. In this thesis, we focused on predicting
three optical properties, namely: i) absorption maximum, ii) emission
maximum, and iii) fluorescence amplification upon binding of the selected
fluorescent molecule to amyloid b fibrils. For this purpose, we employed various
numerical representations of molecules and machine learning methods.
We found that the wavelengths of the absorption and emission maxima negatively
correlate with chemical hardness. Furthermore, feature importance
analysis with the random forest model revealed a substructure consistently
associated with high activity.
|
