Segmentation of Protein Secondary Structure from 3D Cryo-EM Images Using Deep Convolutional Neural Networks
Description/Abstract/Artist Statement
Deep learning is a subset of machine learning that has long been implemented in image processing and classification. One of the popular deep learning networks is deep convolutional neural network (CNN). We have seen applications of deep convolutional neural networks in every aspect of life. Some of them are facial recognition (FaceID, visual search), personalized advertising and health analysis. In parallel with the development of CNNs, cryo-electron microscopy is a Nobel Prize-winning technology that is a critical technique to derive protein structure. Protein structure determines its functionality, thus understanding of the structure is a crucial step to further findings in bioinformatics fields and the search for medicine, vaccines, and health treatments. However, building an accurate structure from 3D Cryo-EM images still remains a challenge for low-resolution images. Multiple deep learning methods have been proposed recently for segmentation of protein secondary structures from 3-dimensional images. One of the goals of this project is to gain understanding of deep learning principles and to apply them in the secondary structure detection problem. Another goal is to evaluate performance of different deep learning methods and identify their strengths and weaknesses.
Faculty Advisor/Mentor
Jing He
College Affiliation
College of Sciences
Presentation Type
Oral Presentation
Disciplines
Computer Sciences
Session Title
Colleges of Sciences UG Research #3
Location
Zoom
Start Date
3-19-2022 3:30 PM
End Date
3-19-2022 4:30 PM
Segmentation of Protein Secondary Structure from 3D Cryo-EM Images Using Deep Convolutional Neural Networks
Zoom
Deep learning is a subset of machine learning that has long been implemented in image processing and classification. One of the popular deep learning networks is deep convolutional neural network (CNN). We have seen applications of deep convolutional neural networks in every aspect of life. Some of them are facial recognition (FaceID, visual search), personalized advertising and health analysis. In parallel with the development of CNNs, cryo-electron microscopy is a Nobel Prize-winning technology that is a critical technique to derive protein structure. Protein structure determines its functionality, thus understanding of the structure is a crucial step to further findings in bioinformatics fields and the search for medicine, vaccines, and health treatments. However, building an accurate structure from 3D Cryo-EM images still remains a challenge for low-resolution images. Multiple deep learning methods have been proposed recently for segmentation of protein secondary structures from 3-dimensional images. One of the goals of this project is to gain understanding of deep learning principles and to apply them in the secondary structure detection problem. Another goal is to evaluate performance of different deep learning methods and identify their strengths and weaknesses.