基于LASSO的多重线性回归在预测血管性痴呆大鼠学习记忆能力中的应用

Application of multiple linear regression based on LASSO in the prediction model of learning and memory ability in vascular dementia rats

目的通过构建基于LASSO的多重线性回归模型,探索预测血管性痴呆(VD)大鼠的学习记忆能力方法。方法采用随机数字表法将40只鼠龄3~5个月,体重(175±25)g的清洁级雄性SD大鼠分为模型组和假手术组,每组20只。采用随机数字表法每组各选5只,用作预测模型验证,其余各15只用作预测模型建立。模型组分次结扎双侧颈总动脉,假手术组分离出颈总动脉后不结扎。采用Morris水迷宫检测大鼠学习记忆能力(潜伏期),造模12周后检测总蛋白、总胆汁酸、总胆固醇等42项血液生化指标,初步建立潜伏期与血液生化指标之间的多重线性回归模型并检测模型是否存在多重共线性,通过LASSO回归筛选血液生化指标预测因子,进一步完善模型,并计算误差率验证模型可行性。结果与假手术组比较,模型组潜伏期显著延长(P<0.05);总蛋白、总胆汁酸、总胆固醇等26项指标的方差膨胀因子数值>10,模型存在多重共线性;LASSO回归筛选出了总胆红素、甘油三酯、胆碱酯酶等10项指标,构建潜伏期与其多重线性回归模型:W=51.887-0.384x_(1)-0.104x_(2)-0.154x_(3)-4.988x_(4)-1157.079x_(5)-7.308x_(6)+7.639x_(7)+0.183x_(8)-0.025x_(9)+34.528x_(10)(x_(1):总蛋白;x_(2):总胆汁酸;x_(3):总胆红素;x_(4):总胆固醇;x_(5):载脂蛋白A;x_(6):高密度脂蛋白胆固醇;x_(7):甘油三酯;x_(8):二氧化碳;x_(9):胆碱酯酶;x_(10):免疫球蛋白M),模型相关系数R=0.852,拟合优度R^(2)=0.725,回归模型统计量值F=5.016,显著性P=0.001,误差率<5%。结论本研究所建立的10项血液生化特征性指标和潜伏期间的数学预测模型,存在明显的线性关系且拟合度较高,不仅可辅助判断实验性VD动物模型复制成功与否,且检测方便快捷,有望为临床VD患者早期预测及干预提供参考。

Objective To explore a method to predict the learning and memory ability of rats with vascular dementia(VD)by constructing a multiple linear regression model based on LASSO.Methods Forty male clean grade SD rats aged three to five months,weighing(175±25)g,were divided into model group and sham operation group by random number table method with 20 rats in each group.Random number table method was used to select five rats from each group for prediction model validation,and 15 rats from each group for prediction model establishment.The model group was prepared by bilateral common carotid artery occlusion in stages.The common carotid artery was not ligated after isolated in the sham operation group.Morris water maze was used to detect the learning and memory ability(latency)of rats.After 12 weeks of modeling,42 blood biochemical indexes such as total protein,total bile acid and total cholesterol were detected.The multiple linear regression model between latency and blood biochemical indexes was initially established and the model was tested for multicollinearity.LASSO regression was used to screen the predictors of blood biochemical indexes to further improve the model,and the error rate was calculated to verify the feasibility of the model.Results Compared with sham operation group,latency of model group was significantly prolonged(P<0.05).Variance inflation factor values of total protein,total bile acid,total cholesterol,and other 26 indexes were more than 10,indicating multicollinearity of the model.LASSO regression screened ten indexes such as total bilirubin,triglyceride,and cholinesterase,and established multiple linear regression model of latency: W = 51.887-0.384x1-0.104x2-0.154x3-4.988x4-1157.079x5-7.308x6+7.639x7+0.183x8-0.025x9+34.528x10 (x1: total protein;x2: total bile acid;x3: total bilirubin;x4: total cholesterol;x5: apolipoprotein A;x6: high density lipoprotein cholesterol;x7: triglyceride;x8: carbon dioxide;x9: cholinesterase;x10: immunoglobulin M) Model correlation coefficient R = 0.852, goodness of fit R^(2) = 0.725, regression model statistical value F = 5.016, significance P = 0.001, error rate was less than 5%. Conclusion The mathematical prediction model of ten blood biochemical characteristics and latency period established in this study has an obvious linear relationship and a high degree of fitting, which can not only assist in judging the success of replication of experimental VD animal model, but also provide a convenient and fast detection, which is expected to provide a reference for early prediction and intervention of clinical VD patients.