基于KG-GCNASL方法的人类癌症合成致死预测研究

Research on Prediction of Cancer Synthetic Lethality Based on KG-GCNASL

癌症合成致死(SL)指两个非必要基因双突变会造成细胞死亡,而两基因分别突变则不会对细胞生长有影响;即在肿瘤细胞中发现一个特定的基因失活会导致肿瘤细胞死亡而正常细胞不会死亡。目前用于此研究的方法包括基因敲除模拟、基于知识数据挖掘与机器学习等,但大多倾向于假设癌症合成致死对象是相互独立的,未考虑潜在的共享生物机制且预测成本较高。因此本文提出了KG-GCNASL方法,基于知识图谱与图卷积网络及注意力机制等相结合的方法解决癌症合成致死预测问题,通过知识图谱所包含的合成致死信息传递合并到预测模型中进行消息的传递与预测,从而规避手动特性工程等问题。本文模型在AUC和F1值等指标已优于目前SL预测其它先进模型,证明该模型对SL预测的显著影响。

Synthetic lethality of cancer(SL)means that double mutation of two unnecessary genes will cause cell death,while separate mutation of two genes will not affect cell growth;that is,a specific gene inactivation found in tumor cells will cause tumor cells to die while normal cells will not.At present,the methods used for this research include gene knockout simulation,knowledge-based data mining and machine learning,but most tend to assume that the synthetic lethal target of cancer is independent of each other,without considering the potential shared biological mechanism and with high prediction cost.Therefore,this paper proposes the KG-GCNASL method,which is based on the combination of knowledge map,graph convolution network and attention mechanism to solve the prediction problem of cancer composite death.Through the transmission of the composite death information contained in the knowledge map and merging it into the prediction model,the message transmission and prediction can be carried out,thus avoiding the problems such as manual characteristic engineering.The AUC and F1 values of the model in this paper are better than other advanced models for SL prediction,which proves that the model has a significant impact on SL prediction.