Applied Basic Research On Constructing Lung Adenocarcinoma Prognosis Model Based On Platelet-inflammatory Factor Characteristics

Part Ⅰ: Predicting Postoperative Prognosis for Patients with Lung Adenocarcinoma Based on Platelet-inflammatory Factors and Machine Learning Objective: Artificial intelligence,especially machine learning,is currently in full swing in medicine,and many studies have reported that machine learning can provide clinical benefits and research value for patients with lung cancer.However,the relationship between machine learning and platelet and lung adenocarcinoma in anesthesia and in the perioperative medical period has rarely been reported.This study explores the relationship between platelet-inflammatory characteristics and lung adenocarcinoma prognostics and provides a predictive model.Methods: We collected preoperative test indexes and clinical variables and constructed prediction models of ultra-long hospitalization days,total postoperative survival and postoperative recurrence of lung adenocarcinoma with 11 machine learning algorithms.We used the GBM and GBDT algorithms to show the first 15 variables that affected the ultra-long hospitalization days and the total postoperative survival and postoperative recurrence of lung adenocarcinoma from high to low.To analyze the relationship between the 15 variables and ultra-long hospitalization days,total postoperative survival and lung adenocarcinoma postoperative recurrence,we used a Pearson correlation.We used accuracy rate,precision rate,ROC curve and other indexes to evaluate the model’s prediction performance.All patients were randomly divided into either a training group or test group at a ratio of 7:3,and we conducted 5-fold cross-validation.Furthermore,we used relevant variables to construct an online web tool for overall survival and postoperative recurrence of lung adenocarcinoma after surgery.Results: The results of correlation analysis between each variable and prognosis were as follows: the GBM algorithm showed that age,NLR,urine volume,neutrophils and PLR were the five factors most affecting ultra-long hospitalization days;tumor stage,thoracoscopy,anesthesia time,smoking,age and platelet count were the six main factors most affecting total lung adenocarcinoma survival;and thoracoscopy,neutrophils,urine volume,platelet width and LMR were the five factors most affecting the postoperative recurrence of lung adenocarcinoma.The results of the 11 algorithms in predicting the ultra-long hospitalization days in the test group were as follows: the AUC values of all machine learning classifiers were less than 0.6,except for Logistic Regression and SVC,and the accuracy of all the algorithms exceeded84%,except the GNB algorithm.The results of the 11 algorithms in predicting total survival after surgery in the test group were as follows: the AUC values of all machine learning classifiers were less than 0.760,except for ADAB,and all of the algorithms’ accuracies exceeded 90%,except GNB.The results of the 11 algorithms in predicting the postoperative recurrence in the test group were as follows: the AUC values of all machine learning classifiers were less than 0.600,with the exceptions of Logistic Regression,Gradient Boosting,SVC,MLPC and ADAB,and all the algorithms had accuracy rates exceeding 90% except for GNB.Finally,based on an artificial intelligence algorithm,we constructed an online web tool to predict the total postoperative survival and postoperative recurrence of lung adenocarcinoma(https://share.streamlit.io/zhouchengmao/streamlit＿app＿gnb＿adab＿os＿re/st＿app＿gnb＿adab＿os＿re.py).Conclusion: A prediction model of ultra-long hospitalization days,total postoperative survival and postoperative recurrence of lung adenocarcinoma can be established based on the characteristics of platelet-inflammatory factors and machine learning.In terms of the comprehensive performance of these 11 machine learning algorithms,the GNB and ADAB machine learning algorithms are the best at predicting ultra-long hospitalization days,total postoperative survival,and postoperative recurrence of lung adenocarcinoma.Moreover,platelet-inflammatory factors are one of the main factors affecting the prognosis of patients.Part Ⅱ: Screening for Platelet-related Diagnostic Genes/ Proteins in Patients with Lung Adenocarcinoma Based on Machine Learning and Multi-omics Objective: The aim of this study was to screen platelet-related genes or proteins to evaluate lung adenocarcinoma prognosis through machine learning and multi-omics so as to provide a target for timely treatment or intervention during anesthesia and the perioperative medical period.This would lay the foundation for improving lung adenocarcinoma diagnosis by combining machine learning and platelet-related variables.Methods: We analyzed platelet-related genes and proteins with the TCGA database and samples’ sequencing data in our hospital.The machine learning algorithm screened out the differentially expressed platelet-related genes and proteins and performed weight analysis of platelet-related genes and proteins.Then,we conducted GO and KEGG enrichment analyses.We verified the expression of these platelet-related genes using several methods.Results: Compared with normal tissues,we identified 99 platelet-related differentially expressed genes,including 43 up-regulated genes and 56 down-regulated genes in lung adenocarcinoma and paracancerous tissues.By screening the ROC curve and the GBM algorithm,we obtained five common platelet-related differentially expressed genes: PECAM1,CLEC3 B,WNT3A,CD36 and GAS6.These five diagnostic genes each had AUC values greater than 0.950.The results of sequencing data from tissue specimens from our hospital and in-vitro cell testing showed that there was a statistically significant difference in the expression of these five genes between the two groups(P < 0.05).For the analysis of platelet-related proteins,we used the GBM algorithm to analyze the relationship between the eight platelet-related differential proteins and stage I lung cancer,and the results showed that the first eight platelet-related differential genes could diagnose stage I lung cancer best.After screening through the ROC curve and GBM algorithm,we obtained two common platelet-related differential proteins: Apolipoprotein A-II and BCHE.Of the eight diagnostic proteins,only Apolipoprotein A-II and BCHE proteins had AUC values reater than 0.800.Conclusion: The platelet-related diagnostic genes(PECAM1,CLEC3 B,WNT3A,CD36 and GAS6)and proteins(Apolipoprotein A-II and BCHE)can be screened from patients with lung cancer based on machine learning and multigroup learning Part Ⅲ:Construction and Verification of a Prognosis Model for Patients with Lung Adenocarcinoma Based on Platelet-related Factors in Multi-omics and Machine Learning Objective: This study aimed to screen platelet-related genes and proteins for lung adenocarcinoma prognosis through machine learning and multi-omics studies so as to provide targets for timely treatment or intervention and further explore the relationship between platelet-related genes or proteins and immunity,as well as other related mechanisms.Methods: We analyzed platelet-related proteins and genes through the TCGA database,TCPA database and sequencing data from tissue samples from our hospital.Next,we screened the differentially expressed platelet-related proteins and genes with a machine learning algorithm.Then,we constructed platelet-related genes and protein risk scores with a machine learning algorithm and a single factor COX proportional hazards model.We verified the model’s effectiveness and the relationship between platelet-related proteins or genes and immune response using several methods,and we also investigated any potential relevant mechanisms.Finally,we verified the effects of the platelet-related genes P2RX1 and TUBA4 A on the biological behavior of lung adenocarcinoma cell line A549 in vitro.Results: For proteomics and prognosis of lung adenocarcinoma,we screened out three platelet-related proteins(CD49B,Fibronectin and FOXM1)related to overallsurvival(OS)by COX risk regression analysis and a machine learning algorithm and constructed a platelet-related protein risk score.The results of the ROC and GBM algorithms showed that platelet-related protein risk score’s performance in predicting OS was comparable to that of the tumor stage.Comparing immune cell infiltration levels between high and low platelet-related protein risk score groups showed that the expression of B cell memory and dendritic cells activated was different.FOXM1 is a core protein.For analysis related to transcriptomics and lung cancer prognosis,we screened out eight platelet-related genes related to OS(P2RX1,TUBA4 A,CLEC3B,HYAL1,COL5A1,TIMP1,CCNA2 and ENTPD2)by single multivariate Cox risk regression analysis and a machine learning algorithm,and thus constructed the platelet-related gene risk score.The results of the ROC and GBM algorithms showed that platelet-related gene risk score’s performance in predicting OS was comparable to that of tumor stage.Then,based on the app Streamlit,we constructed an online web tool for total survival and progress-free lung adenocarcinoma survival,based on the LGBM algorithm(https://share.streamlit.io/zhouchengmao/streamlit＿app＿lgbm＿os＿pfi/st＿app＿lgbm＿os＿pfi.py).There was a strong correlation between platelet-related gene risk score and immune response.Additionally,there were significant differences in immune infiltrating cells such as plasma cells and macrophages M0 between the two groups.Several immune checkpoints,including BTLA and CD276,between the two groups,were statistically different.There were also significant differences in immune function between the two groups,including APC co＿inhibition and MHC class I,and there were significant differences in dysfunction,exclusion and TIDE immune functions between them as well.The in vitro experiment showed that the knockout of P2RX1 and TUBA4A genes could change the proliferation,invasion and migration ability of lung adenocarcinoma cell line A549.Moreover,the knockout of P2RX1 promoted macrophages’ ability to infiltrate lung adenocarcinoma cells.Conclusion: A prognosis model for patients with lung adenocarcinoma can be constructed based on platelet-related proteins or genes in multi-omics and machine learning algorithms,and the model tracks immune response.The knockout of P2RX1 and TUBA4A genes can change lung cancer cell line A549’s proliferation,invasion and migration ability,and the knockout of P2RX1 can promote macrophages’ ability to infiltrate lung adenocarcinoma cells.