Prediction and feature selection for fatal gastrointestinal bleeding recurrence in hospital via machine learning / 中华危重病急救医学

Objective To propose a method of prediction for fatal gastrointestinal bleeding recurrence in hospital and a method of feature selection via machine learning models. Methods 728 digestive tract hemorrhage samples were extracted from the first aid database of PLA General Hospital, and 343 patients among them were diagnosed as fatal gastrointestinal bleeding recurrence in hospital. A total of 64 physiological or laboratory indicators were extracted and screened. Based on the ten-fold cross-validation, Logistic regression, AdaBoost and XGBoost were used for classification prediction and comparison. XGBoost was used to search sequence features, and the key indicators for predicting fatal gastrointestinal bleeding recurrence in hospital were screened out according to the importance of the indicators during training. Results Logistic regression, AdaBoost and XGBoost all get better F1.5 score under each feature input dimension, among which XGBoost had the best effect and the highest score, which was able to identify as many patients as possible who might have fatal gastrointestinal bleeding recurrence in hospital. Through XGBoost iteration results, the Top 30 indicators with high importance for predicting fatal gastrointestinal bleeding recurrence in hospital were ranked. The F1.5 scores of the first 12 key indicators peaked at iteration (0.893), including hemoglobin (Hb), calcium (CA), red blood cell count (RBC), mean platelet volume (MPV), mean erythrocyte hemoglobin concentration (MCH), systolic blood pressure (SBP), platelet count (PLT), magnesium (MG), lymphocyte (LYM), glucose (GLU, blood gas analysis), glucose (GLU, blood biochemistry) and diastolic blood pressure (DBP). Conclusions Logistic regression, AdaBoost and XGBoost could achieve the purpose of early warning for predicting fatal gastrointestinal bleeding recurrence in hospital, and XGBoost is the most suitable. The 12 most important indicators were screened out by sequential forward selection.

Drug-target protein interaction prediction based on AdaBoost algorithm / 生物医学工程学杂志

The drug-target protein interaction prediction can be used for the discovery of new drug effects. Recent studies often focus on the prediction of an independent matrix filling algorithm, which apply a single algorithm to predict the drug-target protein interaction. The single-model matrix-filling algorithms have low accuracy, so it is difficult to obtain satisfactory results in the prediction of drug-target protein interaction. AdaBoost algorithm is a strong multiple classifier combination framework, which is proved by the past researches in classification applications. The drug-target interaction prediction is a matrix filling problem. Therefore, we need to adjust the matrix filling problem to a classification problem before predicting the interaction among drug-target protein. We make full use of the AdaBoost algorithm framework to integrate several weak classifiers to improve performance and make accurate prediction of drug-target protein interaction. Experimental results based on the metric datasets show that our algorithm outperforms the other state-of-the-art approaches and classical methods in accuracy. Our algorithm can overcome the limitations of the single algorithm based on machine learning method, exploit the hidden factors better and improve the accuracy of prediction effectively.

Prediction of protein subcellular locations by ensemble of improved K-nearest neighbor / 生物工程学报

Adaboost algorithm with improved K-nearest neighbor classifiers is proposed to predict protein subcellular locations. Improved K-nearest neighbor classifier uses three sequence feature vectors including amino acid composition, dipeptide and pseudo amino acid composition of protein sequence. K-nearest neighbor uses Blast in classification stage. The overall success rates by the jackknife test on two data sets of CH317 and Gram1253 are 92.4% and 93.1%. Adaboost algorithm with the novel K-nearest neighbor improved by Blast is an effective method for predicting subcellular locations of proteins.