Search

Objective To study whether multimodal brain MRI comprising permeability and perfusion measures coupled with machine learning can predict neurocognitive function in young patients with SLE without neuropsychiatric manifestations. Methods SLE patients and healthy controls (HCs) (≤40 years of age) underwent multimodal structural brain MRI that comprised voxel-based morphometry (VBM), magnetization transfer ratio (MTR) and dynamic contrast-enhanced (DCE) MRI in this cross-sectional study. Neurocognitive function assessed by Automated Neuropsychological Assessment Metrics was reported as the total throughput score (TTS). Olfactory function was assessed. A machine learning–based model (i.e. glmnet) was constructed to predict TTS. Results Thirty SLE patients and 10 HCs were studied. Both groups had comparable VBM, MTR, olfactory bulb volume (OBV), olfactory function and TTS. While after correction for multiple comparisons the uncorrected increase in the blood–brain barrier (BBB) permeability parameters compared with HCs did not remain evident in SLE patients, DCE-MRI perfusion parameters, notably an increase in right amygdala perfusion, was positively correlated with TTS in SLE patients (r  = 0.636, false discovery rate P  < 0.05). A machine learning–trained multimodal MRI model comprising alterations of VBM, MTR, OBV and DCE-MRI parameters mainly in the limbic system regions predicted TTS in SLE patients (r  = 0.644, P  < 0.0005). Conclusion Multimodal brain MRI demonstrated increased right amygdala perfusion that was associated with better neurocognitive performance in young SLE patients without statistically significant BBB leakage and microstructural abnormalities. A machine learning–constructed multimodal model comprising microstructural, perfusion and permeability parameters accurately predicted neurocognitive performance in SLE patients. [ABSTRACT FROM AUTHOR]

The trade-off between a machine learning (ML) and deep learning (DL) model's predictability and its interpretability has been a rising concern in central nervous system-related quantitative structure–activity relationship (CNS-QSAR) analysis. Many state-of-the-art predictive modeling failed to provide structural insights due to their black box-like nature. Lack of interpretability and further to provide easy simple rules would be challenging for CNS-QSAR models. To address these issues, we develop a protocol to combine the power of ML and DL to generate a set of simple rules that are easy to interpret with high prediction power. A data set of 940 market drugs (315 CNS-active, 625 CNS-inactive) with support vector machine and graph convolutional network algorithms were used. Individual ML/DL modeling methods were also constructed for comparison. The performance of these models was evaluated using an additional external dataset of 117 market drugs (42 CNS-active, 75 CNS-inactive). Fingerprint-split validation was adopted to ensure model stringency and generalizability. The resulting novel hybrid ensemble model outperformed other constituent traditional QSAR models with an accuracy of 0.96 and an F1 score of 0.95. With the power of the interpretability provided with this protocol, our model laid down a set of simple physicochemical rules to determine whether a compound can be a CNS drug using six sub-structural features. These rules displayed higher classification ability than classical guidelines, with higher specificity and more mechanistic insights than just for blood–brain barrier permeability. This hybrid protocol can potentially be used for other drug property predictions. [ABSTRACT FROM AUTHOR]

Motivation Identification of blood–brain barrier (BBB) permeability of a compound is a major challenge in neurotherapeutic drug discovery. Conventional approaches for BBB permeability measurement are expensive, time-consuming and labor-intensive. BBB permeability is associated with diverse chemical properties of compounds. However, BBB permeability prediction models have been developed using small datasets and limited features, which are usually not practical due to their low coverage of chemical diversity of compounds. Aim of this study is to develop a BBB permeability prediction model using a large dataset for practical applications. This model can be used for facilitated compound screening in the early stage of brain drug discovery. Results A dataset of 7162 compounds with BBB permeability (5453 BBB+ and 1709 BBB-) was compiled from the literature, where BBB+ and BBB- denote BBB-permeable and non-permeable compounds, respectively. We trained a machine learning model based on Light Gradient Boosting Machine (LightGBM) algorithm and achieved an overall accuracy of 89%, an area under the curve (AUC) of 0.93, specificity of 0.77 and sensitivity of 0.93, when 10-fold cross-validation was performed. The model was further evaluated using 74 central nerve system compounds (39 BBB+ and 35 BBB-) obtained from the literature and showed an accuracy of 90%, sensitivity of 0.85 and specificity of 0.94. Our model outperforms over existing BBB permeability prediction models. Availabilityand implementation The prediction server is available at http://ssbio.cau.ac.kr/software/bbb. [ABSTRACT FROM AUTHOR]

Objectives: The aim of this study was to use Gaussian process regression (GPR) methods to quantify the effect of experimental temperature (Texp) and choice of diffusion cell on model quality and performance. Methods: Data were collated from the literature. Static and flow‐through diffusion cell data were separated, and a series of GPR experiments was conducted. The effect of Texp was assessed by comparing a range of datasets where Texp either remained constant or was varied from 22 to 45 °C. Key findings: Using data from flow‐through diffusion cells results in poor model performance. Data from static diffusion cells resulted in significantly greater performance. Inclusion of data from flow‐through cell experiments reduces overall model quality. Consideration of Texp improves model quality when the dataset used exhibits a wide range of experimental temperatures. Conclusions: This study highlights the problem of collating literature data into datasets from which models are constructed without consideration of the nature of those data. In order to optimise model quality data from only static, Franz‐type, experiments should be used to construct the model and Texp should either be incorporated as a descriptor in the model if data are collated from a range of studies conducted at different temperatures. [ABSTRACT FROM AUTHOR]

Abstract: Objectives: The aim of this study was to investigate how to improve predictions from Gaussian Process models by optimising the model hyperparameters. Methods: Optimisation methods, including Grid Search, Conjugate Gradient, Random Search, Evolutionary Algorithm and Hyper‐prior, were evaluated and applied to previously published data. Data sets were also altered in a structured manner to reduce their size, which retained the range, or ‘chemical space’ of the key descriptors to assess the effect of the data range on model quality. Key findings: The Hyper‐prior Smoothbox kernel results in the best models for the majority of data sets, and they exhibited significantly better performance than benchmark quantitative structure–permeability relationship (QSPR) models. When the data sets were systematically reduced in size, the different optimisation methods generally retained their statistical quality, whereas benchmark QSPR models performed poorly. Conclusions: The design of the data set, and possibly also the approach to validation of the model, is critical in the development of improved models. The size of the data set, if carefully controlled, was not generally a significant factor for these models and that models of excellent statistical quality could be produced from substantially smaller data sets. [ABSTRACT FROM AUTHOR]