Blood-Brain-Barrier (BBB) is a strict permeability barrier for maintaining the Central Nervous System (CNS) homeostasis. One of the most important conditions to judge a CNS drug is to figure out whether it has BBB permeability or not. In the past 20 years, the existing prediction approaches are usually based on the data of the physical characteristics and chemical structure of drugs. However, these methods are usually only applicable to small molecule compounds based on passive diffusion through BBB. To deal this problem, one of the most famous methods is multi-core SVM method, which is based on clinical phenotypes about Drug Side Effects and Drug Indications to predict drug penetration of BBB. This paper proposed a Deep Learning method to predict the Blood-Brain-Barrier permeability based on the clinical phenotypes data. The validation result on three datasets proved that Deep Learning method achieves better performance than the other existing methods. The average accuracy of our method reaches 0.97, AUC reaches 0.98, and the F1 score is 0.92. The results proved that Deep Learning methods can significantly improve the prediction accuracy of drug BBB permeability and it can help researchers to reduce clinical trials and find new CNS drugs.

Identifying drug–target interactions (DTIs) plays an important role in the field of drug discovery, drug side-effects, and drug repositioning. However, in vivo or biochemical experimental methods for identifying new DTIs are extremely expensive and time-consuming. Recently, in silico or various computational methods have been developed for DTI prediction, such as ligand-based approaches and docking approaches, but these traditional computational methods have several limitations. This work utilizes the chemogenomic-based approaches for efficiently identifying potential DTI candidates, namely, self-paced learning with collaborative matrix factorization based on weighted low-rank approximation (SPLCMF) for DTI prediction, which integrates multiple networks related to drugs and targets into regularized least-squares and focuses on learning a low-dimensional vector representation of features. The SPLCMF framework can select samples from easy to complex into training by using soft weighting, which is inclined to more faithfully reflect the latent importance of samples in training. Experimental results on synthetic data and five benchmark data sets show that our proposed SPLCMF outperforms other existing state-of-the-art approaches. These results indicate that our proposed SPLCMF can provide a useful tool to predict unknown DTIs, which may provide new insights into drug discovery, drug side-effect prediction, and repositioning existing drug.

Abstract. Tuberculosis (TB), caused by infection with mycobacterium tuberculosis, is still a major threat to human health worldwide. Current diagnostic methods encounter some limitations, such as sample collection problem or unsatisfied sensitivity and specificity issue. Moreover, it is hard to identify TB from some of other lung diseases without invasive biopsy. In this paper, the logistic models with three representative regularization approaches including Lasso (the most popular regularization method), and ‫ܮ‬ ଵ/ଶ (the method that inclines to achieve more sparse solution than Lasso) and Elastic Net (the method that encourages a grouping effect of genes in the results) adopted together to select the common gene signatures in microarray data of peripheral blood cells. As the result, 13 common gene signatures were selected, and sequentially the classifier based on them is constructed by the SVM approach, which can accurately distinguish tuberculosis from other pulmonary diseases and healthy controls. In the test and validation datasets of the blood gene expression profiles, the generated classification model achieved 91.86% sensitivity and 93.48% specificity averagely. Its sensitivity is improved 6%, but only 26% gene signatures used compared to recent research results. These 13 gene signatures selected by our methods can be used as the basis of a blood-based test for the detection of TB from other pulmonary diseases and healthy controls.

Gene selection is an attractive and important task in cancer survival analysis. Most existing supervised learning methods can only use the labeled biological data, while the censored data (weakly labeled data) far more than the labeled data are ignored in model building. Trying to utilize such information in the censored data, a semi-supervised learning framework (Cox-AFT model) combined with Cox proportional hazard (Cox) and accelerated failure time (AFT) model was used in cancer research, which has better performance than the single Cox or AFT model. This method, however, is easily affected by noise. To alleviate this problem, in this paper we combine the Cox-AFT model with self-paced learning (SPL) method to more effectively employ the information in the censored data in a self-learning way. SPL is a kind of reliable and stable learning mechanism, which is recently proposed for simulating the human learning process to help the AFT model automatically identify and include samples of high confidence into training, minimizing interference from high noise. Utilizing the SPL method produces two direct advantages: (1) The utilization of censored data is further promoted; (2) the noise delivered to the model is greatly decreased. The experimental results demonstrate the effectiveness of the proposed model compared to the traditional Cox-AFT model.

The quantitative structure-activity relationship (QSAR) model searches for a reliable relationship between the chemical structure and biological activities in the field of drug design and discovery. (1) Background: In the study of QSAR, the chemical structures of compounds are encoded by a substantial number of descriptors. Some redundant, noisy and irrelevant descriptors result in a side-effect for the QSAR model. Meanwhile, too many descriptors can result in overfitting or low correlation between chemical structure and biological bioactivity. (2) Methods: We use novel log-sum regularization to select quite a few descriptors that are relevant to biological activities. In addition, a coordinate descent algorithm, which uses novel univariate log-sum thresholding for updating the estimated coefficients, has been developed for the QSAR model. (3) Results: Experimental results on artificial and four QSAR datasets demonstrate that our proposed log-sum method has good performance among state-of-the-art methods. (4) Conclusions: Our proposed multiple linear regression with log-sum penalty is an effective technique for both descriptor selection and prediction of biological activity.