基于多组学数据的肺癌分期预测研究

Study on Multi-Omics Data-Driven Prediction of Lung Cancer Stages

摘要: 癌症是一类由基因变异引发的恶性疾病，其发病率和死亡率均较高，严重威胁人类健康。基因表达调控对生物体发育至关重要，在肿瘤发生与发展中，常表现为沉默基因的异常激活或活跃基因的表达抑制，这被认为是促进肿瘤发展的关键机制之一。此外，人体微生物群落参与调控多种生理过程，其结构或功能失调可提升致癌风险。本研究聚焦于肺癌，整合基因表达与微生物组数据，旨在开发一种用于肿瘤分期预测的计算模型。研究流程如下：首先对基因表达与微生物组数据进行差异分析，筛选显著变化的基因及微生物物种；其次构建融合注意力机制的深度神经网络模型；随后基于弹性网模型选出的关键特征训练模型以预测肺癌分期；最后采用五折交叉验证评估模型性能。实验结果表明，该模型在肺癌分期预测中表现优异，准确率超过80%。

Abstract: Cancer is a malignant disease driven by genetic alterations, characterized by high incidence and mortality rates, posing a severe threat to human health. Precise regulation of gene expression is essential for normal organismal development; in tumorigenesis and progression, it is frequently disrupted through aberrant activation of normally silenced genes or suppression of constitutively active genes—a mechanism widely regarded as pivotal in cancer development. Moreover, the human microbiota modulates a wide array of physiological processes, and dysbiosis—either structural or functional—has been associated with an elevated risk of carcinogenesis. This study focuses on lung cancer and integrates gene expression and microbiome data to develop a computational model for tumor stage prediction. The workflow is as follows: To identify significantly dysregulated genes and microbial taxa, differential analyses were conducted on both gene expression profiles and microbiome compositions. Subsequently, a deep neural network incorporating an attention mechanism is constructed; third, key features selected by an elastic net model are used to train the network for lung cancer staging; Finally, model performance is evaluated via five-fold cross-validation. Experimental results demonstrate that the proposed model achieves superior predictive performance, with an accuracy exceeding 80%.