(6b) A Novel Computer-Aided Molecular Approach Using the Signature Molecular Descriptor to Design Potentially New Non-Intuitive Amyloid-? Aggregation Inhibitors

The ‘amyloid cascade hypothesis’, linking aggregation of the amyloid-β protein (Aβ) to the pathogenesis of Alzheimer’s disease (AD), has led to the emergence of inhibition of Aβ aggregation as a therapeutic strategy for this currently unpreventable and devastating disease.  The identification of Aβ aggregation inhibitors has proceeded primarily through the screening of drug-like molecules, leading to the discovery of hundreds of aggregation inhibitors belonging to multiple classes of compounds.  However, only two small-molecule amyloid aggregation inhibitors have proven effective enough to advance to late clinical trials. Albeit proven effective, identifying drugs from specific classes with optimal inhibition effect can be incremental, time consuming, and expensive where improved inhibitors are often developed by small “refinements”. Such an approach cannot identify novel and non-intuitive inhibitors with desired properties, however.

Alternatively, a computer-aided molecular design (CAMD) strategy that uses an inverse quantitative structure property relationship (I-QSPR) is proposed for the design of drug-like molecules. The CAMD technique is a powerful tool that applies information about how known substances behave in a given application, encodes their molecular structure using what are referred to as “molecular descriptors,” relates the descriptors to its performance, then uses that information to generate new molecules with optimal predicted performance without the need for detailed information about drug target. Such an approach has the potential to accelerate the discovery of drug-like molecules. 

In this work, 21 molecules that belong to the classes “proven effective” including naphthalimides, coumarin analogs, dihydropyridines, and polyphenols were evaluated for their effect upon aggregation of Aβ monomer. Aggregation was initiated by incubating monomeric protein in the presence of NaCl and under continuous agitation.  Progression from monomer to aggregates was evaluated using thioflavin T, a fluorescent dye that yields a shifted and enhanced fluorescence when bound to the β-sheet structure of amyloid aggregates.  Uninhibited aggregation exhibited a lag time, characteristic of nucleation, followed by a period of rapid growth, and culminating with a plateau as monomer and aggregate species reached equilibrium.  Compounds were characterized for both their ability to extend nucleation, indicated by a fold-increase in the lag time, and their ability to reduce the quantity of aggregate formed, indicated by a decrease in the fluorescence of the equilibrium plateau. After deconstructing all evaluated structures into their corresponding building blocks, or molecular descriptors, two quantitative structure property relationship (QSPR) models were created and refined in order to score the over 200 million inverse solutions, which are potential structures. Both QSPR models employed advanced machine-learning techniques. For the lag extension time, a classifier was developed where solutions with a positive lag extension time greater than 2-fold were kept. For the decrease in equilibrium plateau, a support vector regression model was created to filter out solutions with percent decreases less than 95%. The final step of refinement ensured drug-likeness of model outputs where solutions were screened via Lipinski’s Rule.  Optimal inverse solutions were then converted into structures by employing an enumeration algorithm and structures with a certain molecular stability were retained. The next step was to identify commercially available compounds or synthesize optimal candidates and test them for efficacy. Screening the PubChem chemical database in order to identify structures similar to the CAMD output based on a specific predefined similarity metric could expedite the discovery of potentially new drugs if structure synthesis was inevitable.