Evaluation of a Deep Learning Method for Protein Secondary Structure Segmentation from Cryo-Electron Microscopy Density Maps at Medium Resolution
Description/Abstract/Artist Statement
A protein’s atomic structure determines its function. Understanding the structure and function of a protein is fundamental in several different applications, including but not limited to drug design, disease treatment and diagnosis, and advancing biotechnology. Cryo-electron microscopy (cryo-EM) is an imaging process yielding three-dimensional density maps of proteins and other molecules that can be used to derive their atomic structures. While highly effective in high resolution, particular challenges arise when using a cryo-EM density map in medium resolution (5-10Å) to derive a protein’s atomic structure. In such cases, the segmentation of protein secondary structures (helices and β sheets) provides constraints for deriving the atomic structure. DeepSSETracer is a software bundle within ChimeraX which aims to accomplish that task. It is a deep learning model created using a U-net convolutional neural network architecture to detect protein secondary structures in medium-resolution cryo-EM density maps. A newly created ad hoc dataset will be combined with randomly sampled existing data to test the performance of DeepSSETracer in its segmentation of protein secondary structures. Performance will be determined using a program to calculate an F1 score by comparing the predicted secondary structures with known solution structures archived in the Protein Data Bank.
Faculty Advisor/Mentor
Jing He
Faculty Advisor/Mentor Department
Computer Science
College Affiliation
College of Sciences
Presentation Type
Oral Presentation
Disciplines
Artificial Intelligence and Robotics | Bioinformatics | Data Science
Session Title
College of Sciences 4
Location
Learning Commons @Perry Library, Room 1310
Start Date
3-30-2024 12:00 PM
End Date
3-30-2024 1:00 PM
Evaluation of a Deep Learning Method for Protein Secondary Structure Segmentation from Cryo-Electron Microscopy Density Maps at Medium Resolution
Learning Commons @Perry Library, Room 1310
A protein’s atomic structure determines its function. Understanding the structure and function of a protein is fundamental in several different applications, including but not limited to drug design, disease treatment and diagnosis, and advancing biotechnology. Cryo-electron microscopy (cryo-EM) is an imaging process yielding three-dimensional density maps of proteins and other molecules that can be used to derive their atomic structures. While highly effective in high resolution, particular challenges arise when using a cryo-EM density map in medium resolution (5-10Å) to derive a protein’s atomic structure. In such cases, the segmentation of protein secondary structures (helices and β sheets) provides constraints for deriving the atomic structure. DeepSSETracer is a software bundle within ChimeraX which aims to accomplish that task. It is a deep learning model created using a U-net convolutional neural network architecture to detect protein secondary structures in medium-resolution cryo-EM density maps. A newly created ad hoc dataset will be combined with randomly sampled existing data to test the performance of DeepSSETracer in its segmentation of protein secondary structures. Performance will be determined using a program to calculate an F1 score by comparing the predicted secondary structures with known solution structures archived in the Protein Data Bank.