Application of multilayered strategy for variable selection in QSAR modeling of PET and SPECT imaging agents as diagnostic agents for Alzheimer's disease

Non-invasive imaging of amyloid beta (A beta) and tau fibrils in the brain may support an early and precise diagnosis of Alzheimer's disease. Molecular imaging technologies involving radionuclides such as positron emission tomography (PET) and single-photon emission computed tomography (SPECT) against beta amyloid plaques and tau fibrils are among emerging research areas in the field of medicinal chemistry. In the current study, we have developed partial least square (PLS) regression-based two-dimensional quantitative structure-activity relationship (2D-QSAR) models using datasets of 38 PET and 73 SPECT imaging agents targeted against A beta protein and 31 imaging agents (both PET and SPECT) targeted against tau protein. Following the strict Organization for Economic Co-operation and Development (OECD) guidelines, we have strived to select significant descriptors from the large initial pool of descriptors using multilayered variable selection strategy using the double cross-validation (DCV) method followed by the best subset selection (BSS) method prior to the development of the final PLS models. The developed models showed significant statistical performance and reliability. Molecular docking studies have been performed to understand the molecular interactions between the ligand and receptor, and the results are then correlated with the structural features obtained from the QSAR models. Furthermore, we have also designed some imaging agents based on the information provided by the models developed and some of them are predicted to be similar to or more active than the most active imaging agents present in the original dataset.