Prediction of DNA Profiles Using STR Electrophoresis Images Based on Convolutional Neural Network

Abstract

The convolutional neural network (CNN) is a deep learning method to recognize images with high accuracy and low error rate. In this study, we performed classification of DNA profiles based on electropherogram images using the CNN. 1,800 DNA profiles images of three control DNAs (007, 2800M, 9947A) were used for CNN model as a dataset. A dataset was divided into the train and test data set, and 1,500 and 300 images were used, respectively. The CNN model used the LeNet-5 architecture, and an accuracy of model was estimated by k-fold cross validation. Both the training data set and test data set showed an accuracy of 1.0 and a loss rate of 0 with 50 epochs. All 300 images of test data set exactly matched the actual data, and the predict probability of match for each element showed 0.579.

Related Papers

• → A Large‐Scale Empirical Evaluation of Cross‐Validation and External Test Set Validation in (Q)SAR(2013)61 cited
• → What size test set gives good error rate estimates?(1998)153 cited
• → Investigations to Minimum Phone Error Training in Bilingual Speech Recognition(2009)6 cited
• → Reliable estimation of externally validated prediction errors for QSAR models(2013)1 cited
• → Prediction of DNA Profiles Using STR Electrophoresis Images Based on Convolutional Neural Network(2022)