材料科学
人工智能
曲线拟合
计算机科学
熔化曲线分析
模式识别(心理学)
生物系统
数学
DNA
特征(语言学)
作者
Zheng Wang,Jingzhen Li,Bin Liu,Yayun Chen Ning Lin,Wangming XU,Zedong Nie
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2025-12-12
卷期号:: 2164-2168
标识
DOI:10.1109/iccc68654.2025.11437661
摘要
A critical bottleneck in genotyping via High-Resolution Melting (HRM) analysis lies in the subjectivity and labor intensity of manual curve interpretation, whereas traditional machine learning approaches depend on suboptimal, manually designed feature extraction. To address these challenges, this study develops an automated, end-to-end deep learning framework for the direct classification of HRM data and proposes an improved ResNet-20 model. Benefiting from the integration of a residual structure, the hierarchical architecture of the Convolutional Neural Network (CNN) employed in this model allows for increased depth. Applied to melting curve processing, this model enables the automatic learning of deep discriminative morphological features. The proposed model was trained and validated on 168 DNA samples to discriminate between wild-type and mutant genotypes. Experimental results show that the improved ResNet-20 model attains state-of-the-art (SOTA) performance, achieving an accuracy of 0.941 and an Area Under the Curve (AUC) of 0.996. This performance demonstrates its exceptional capability to capture subtle and complex patterns in melting curve profiles. By directly learning features from raw HRM data, the proposed deep learning framework circumvents the limitations of manual feature engineering and establishes a reliable, objective, and data-driven HRM analysis approach. This study substantially enhances the accuracy and efficiency of genotyping based on the human genome, thereby offering a highly promising tool for genetic screening applications.
科研通智能强力驱动
Strongly Powered by AbleSci AI