METHODS: We followed up 240 participants (112 cognitively unimpaired [CU], 78 amnestic mild cognitive impairment [aMCI], and 50 Alzheimer's disease (AD) dementia [ADD]) for 2 years from 9 referral centers in South Korea. Participants were assessed with neuropsychological tests and 18F-flutemetamol (FMM) positron emission tomography (PET). Ten regions (frontal, precuneus/posterior cingulate (PPC), lateral temporal, parietal, and striatum of each hemisphere) were visually examined in the FMM scan, and participants were divided into three groups: No-FMM, Focal-FMM (FMM uptake in 1-9 regions), and Diffuse-FMM. We used mixed-effects model to investigate the speed of cognitive decline in the Focal-FMM group according to the cognitive level, extent, and location of Aß involvement, in comparison with the No- or Diffuse-FMM group.
RESULTS: Forty-five of 240 (18.8%) individuals were categorized as Focal-FMM. The rate of cognitive decline in the Focal-FMM group was faster than the No-FMM group (especially in the CU and aMCI stage) and slower than the Diffuse-FMM group (in particular in the CU stage). Within the Focal-FMM group, participants with FMM uptake to a larger extent (7-9 regions) showed faster cognitive decline compared to those with uptake to a smaller extent (1-3 or 4-6 regions). The Focal-FMM group was found to have faster cognitive decline in comparison with the No-FMM when there was uptake in the PPC, striatum, and frontal cortex.
CONCLUSIONS: When predicting cognitive decline of patients with focal Aß deposition, the patients' cognitive level, extent, and location of the focal involvement are important.
RESULTS: In this study, we established an ensemble computational model for discriminating the near-native docking pose of APKs named "AnkPlex". A dataset of APKs was generated from seven X-ray APKs, which consisted of 3 internal domains, using the reliable docking tool ZDOCK. The dataset was composed of 669 and 44,334 near-native and non-near-native poses, respectively, and it was used to generate eleven informative features. Subsequently, a re-scoring rank was generated by AnkPlex using a combination of a decision tree algorithm and logistic regression. AnkPlex achieved superior efficiency with ≥1 near-native complexes in the 10 top-rankings for nine X-ray complexes compared to ZDOCK, which only obtained six X-ray complexes. In addition, feature analysis demonstrated that the van der Waals feature was the dominant near-native pose out of the potential ankyrin-protein docking poses.
CONCLUSION: The AnkPlex model achieved a success at predicting near-native docking poses and led to the discovery of informative characteristics that could further improve our understanding of the ankyrin-protein complex. Our computational study could be useful for predicting the near-native poses of binding proteins and desired targets, especially for ankyrin-protein complexes. The AnkPlex web server is freely accessible at http://ankplex.ams.cmu.ac.th .