基于孪生网络的抗乳腺癌药物活性筛选研究

打开文本图片集

中图分类号：TP18 文献标识码：A 文章编号：2096-4706（2026）04-0138-09

Research on Anti-breast Cancer Drug Activity Screening Based on Siamese Network

LI Yangfei1-2，， WANG Qiumeng13，DAI Huarong4 （1.School ofArtificial Intelligence，Mianyang Polytechnic，Mianyang 621ooo， China; 2.College ofLife Science，China WestNormal University，Nanchong637oo9，China; yangEngineering TechnologicalResearch Centerof Visual ObjectDetectionandRecognition，Mianyang 6210oo，China; 4.Department of Thyroid Surgery，Mianyang Third People's Hospital，Mianyang 621ooo， China）

Abstract： Breast cancer isone of the most common malignant tumors in women，and precise and eficient screening methodsforcandidatedrugsareurgently needed.This paper proposesa three-layer screening framework basedon Siamese networks.Fistly，itflterskeymolecularescriptors troughK-MeansclusteringandMutualIformationaalysistoalleviate high-dimensional redundancy. Secondly， dual-branch multi-layer perceptrons are adopted to predict IC50nM and pIC50 （20 respectivelyandtheiamesenetwork isutilized toenhancesmallsampleADMETdata toimproveclassficationperformance. Finaly，itrealizesmultbjectiveotimizationbyombiningGeneticAgorithmsandDecisionTreeules.Experimentalresults show that the model reduces the activity prediction error by 15.2% ，and the ADMET classification accuracy increases to 87.5% The framework successfully screens 632 potential candidate molecules with both activity and drug-likenes，providing strong computational support for the early research and development of anti-breast cancer drugs.

Keywords： breast cancer; Estrogen Receptors a （ERα）; siamese network; bioactivity prediction; ADMET property optimization; candidate drug screening

0 引言

乳腺癌作为一种典型的激素依赖性肿瘤，具有较高的女性发病率和致死率，其病理机制与雌激素受体（Estrogen Receptors，ER）的异常表达密切相关[]。（剩余12914字）