Class-imbalance Prediction and High-dimensional Risk Factor Identification of Adverse Reactions of Traditional Chinese Medicine with Centralized Monitoring in Real-world Hospitals / 中国实验方剂学杂志

ObjectiveTo achieve high-dimensional prediction of class imbalanced of adverse drug reaction（ADR） of traditional Chinese medicine（TCM） and to classify and identify risk factors affecting the occurrence of ADR based on the post-marketing safety data of TCM monitored centrally in real world hospitals. MethodThe ensemble clustering resampling combined with regularized Group Lasso regression was used to perform high-dimensional balancing of ADR class-imbalanced data， and then to integrate the balanced datasets to achieve ADR prediction and the risk factor identification by category. ResultA practical example study of the proposed method on a monitoring data of TCM injection performed that the accuracy of the ADR prediction， the prediction sensitivity， the prediction specificity and the area under receiver operating characteristic curve（AUC） were all above 0.8 on the test set. Meanwhile， 40 risk factors affecting the occurrence of ADR were screened out from total 600 high-dimensional variables. And the effect of risk factors on the occurrence of ADR was identified by classification weighting. The important risk factors were classified as follows：past history， medication information， name of combined drugs， disease status， number of combined drugs and personal data. ConclusionIn the real world data of rare ADR with a large amount of clinical variables， this paper realized accurate ADR prediction on high-dimensional and class imbalanced condition， and classified and identified the key risk factors and their clinical significance of categories， so as to provide risk early warning for clinical rational drug use and combined drug use， as well as scientific basis for reevaluation of safety of post-marketing TCM.

Single-channel electroencephalogram signal used for sleep state recognition based on one-dimensional width kernel convolutional neural networks and long-short-term memory networks / 生物医学工程学杂志

Aiming at the problem that the unbalanced distribution of data in sleep electroencephalogram(EEG) signals and poor comfort in the process of polysomnography information collection will reduce the model's classification ability, this paper proposed a sleep state recognition method using single-channel EEG signals (WKCNN-LSTM) based on one-dimensional width kernel convolutional neural networks(WKCNN) and long-short-term memory networks (LSTM). Firstly, the wavelet denoising and synthetic minority over-sampling technique-Tomek link (SMOTE-Tomek) algorithm were used to preprocess the original sleep EEG signals. Secondly, one-dimensional sleep EEG signals were used as the input of the model, and WKCNN was used to extract frequency-domain features and suppress high-frequency noise. Then, the LSTM layer was used to learn the time-domain features. Finally, normalized exponential function was used on the full connection layer to realize sleep state. The experimental results showed that the classification accuracy of the one-dimensional WKCNN-LSTM model was 91.80% in this paper, which was better than that of similar studies in recent years, and the model had good generalization ability. This study improved classification accuracy of single-channel sleep EEG signals that can be easily utilized in portable sleep monitoring devices.

Improving the Prediction of Protein-Protein Interaction Sites Using a Novel Over-Sampling Approach and Predicted Shape Strings.

Identification of protein-protein interaction (PPI) sites is one of the most challenging tasks in bioinformatics and many computational methods based on support vector machines have been developed. However, current methods often fail to predict PPI sites mainly because of the severe imbalance between the numbers of interface and non-interface residues. In this study, we propose a novel over-sampling method that relaxes the class-imbalance problem based on local density distributions. We applied the proposed method to a PPI dataset that includes 2,829 interface and 24,616 non-interface residues. The experimental result showed a significant improvement in predictive performance comparing with the other state-of-the-art methods according to the six evaluation measures.