Research On Key Technologies Of Intelligent Drug Redirection And Disease-Assisted Decision

In recent years,the field of smart healthcare has made important breakthroughs under the influence of the development of deep learning.The importance of research and development of innovative drugs and intelligent assisted medical technology,which are closely related to people’s lives,has been widely recognized.Among them,smart drug redirection and disease-assisted decision-making complement each other.On the one hand,it explores "new use of old drugs".On the other hand,it simultaneously realizes the value of clinical transformation.From drug discovery to clinical application,it creates an integrated closed-loop ecosystem of smart medicine.Specifically,this article starts from the two directions of drug target targeting prediction and rare disease-assisted decision-making.Aiming at the problems of low prediction accuracy and poor interpretability,deep learning methods are used to improve.In addition,this article has also completed the design and implementation of the clinical auxiliary diagnosis system,which can effectively improve the diagnosis efficiency of doctors.The specific research content of this paper is as follows:1)A targeting prediction model based on sub-structure representation is proposed.The model first enriches the characteristic information of the drug and target by fusing the heterogeneous graph.Secondly,a mutual information estimator based on sub-structure is designed.On the one hand,the noise information in the graph network is effectively eliminated.On the other hand,the correlation between the graph-level representation and the sub-graph representation is enhanced.Finally,based on the high-quality embedding of the drugs and targets,an end-to-end self-decoder model is proposed to complete the downstream task of link prediction.Case studies and comparative experiments also show the effectiveness of the model.2)A rare disease-assisted decision-making method based on RNNs enhanced graph network model is proposed.The model infers the probability of rare diseases based on the patient’s main complaint.The text structuring module is based on the pre-training model of BERT to capture the overall syntax of the text.The CNN model is used to model the local semantics to identify the patient’s intention and word slot.The RNNs model is designed in the disease-assisted decision-making module to capture the information of multi-hop nodes around rare diseases.A random walk algorithm with weights is proposed to avoid blindness in the walking process.Finally we transform it into a language model problem for prediction.In the experiment,the model achieved the highest scores on the real disease graph,which demonstrates the advancement of the model.3)Based on the disease prediction model in the previous part,a clinical decision-making system is constructed.The system adopts an active interactive method to give the possibility of suffering from the disease during the diagnosis process,and list the cause of the disease,treatment plan,and precautions.At the same time,the system uses Elasticsearch technology to assist doctors in quickly retrieving symptom information.