基于机器学习的前列腺癌致病基因筛选与疾病预测

Screening and Disease Prediction of Prostate Cancer-Causing Genes Based on Machine Learning

摘要: 为识别与前列腺癌相关的遗传特征，文章提出了一种集成式机器学习方法，用于筛选前列腺癌的关键基因，并深入分析这些靶基因的生物学意义，从而建立高效的疾病诊断预测模型。研究通过UCSC Xena数据库收集了151例前列腺癌组织和152例正常组织的转录组数据，并采用PCA等方法进行批次效应校正。通过差异表达分析筛选出了2586个上下调基因，并结合GO和KEGG富集分析，揭示了与前列腺癌相关的关键致病通路。进一步集成随机森林、LASSO回归和梯度提升机(GBM)三种机器学习算法进行基因二次筛选，最终确定了12个对前列腺癌具有重要影响的关键基因。基于这些基因，构建了8种前列腺癌诊断预测模型，采用混淆矩阵和ROC曲线对模型性能进行评估。结果显示，极限梯度提升(XGBoost)模型的准确度和AUC值分别达到了93%和97%，验证了该模型在前列腺癌诊断中的应用潜力。

Abstract: To identify genetic features associated with prostate cancer, the article proposes an integrated machine learning approach for screening key genes for prostate cancer and analyzing the biological significance of these target genes in-depth to build an efficient predictive model for disease diagnosis. The study collected transcriptome data from 151 prostate cancer tissues and 152 normal tissues through the UCSC Xena database, and corrected for batch effects using PCA and other methods. A total of 2586 up- and down-regulated genes were screened by differential expression analysis and combined with GO and KEGG enrichment analysis to reveal the key pathogenic pathways associated with prostate cancer. A secondary screening was further integrated with three machine learning algorithms, namely Random Forest, LASSO regression and Gradient Boosting Machine (GBM), and 12 key genes with significant impact on prostate cancer were finally identified. Based on these genes, eight prostate cancer diagnosis prediction models were constructed, and the model performance was evaluated using confusion matrix and ROC curve. The results showed that the accuracy and AUC value of the Extreme Gradient Boosting (XGBoost) model reached 93% and 97%, respectively, which verified the potential application of the model in prostate cancer diagnosis.