Machine Learning-Guided Design of Novel Adeno-Associated Virus (AAV) Vectors for Human Gene Therapy

Overview

Gene therapy has immense potential for treating conditions such as cancer, hemophilia, kidney disease, and inherited blindness by correcting defective genes directly. Adeno-associated viruses (AAV) are commonly used in gene therapy development and research due to their non-pathogenic nature in humans. A major bottleneck in designing safer and more efficient gene therapy treatments lies in creating novel viral vectors that exhibit reduced immune response or tissue specificity. Recent efforts have involved using machine learning to overcome this bottleneck by identifying novel viral vectors for gene therapy. Employing machine learning to screen out viral vectors that are predicted to be non-viable accelerates benchwork by reducing the number of mutant viruses that must be experimentally characterized for downstream application. In this project, we created a deep convolutional neural network (CNN) model that incorporates natural language processing (NLP) techniques to determine AAV capsid viability from its primary amino-acid sequence. Specifically, we created and trained a CNN model on the C1R2 dataset of complete single point-mutations plus random double mutations described in the paper “Deep diversification of an AAV capsid protein by machine learning” by Bryant et al. (2021). Our model achieved an accuracy of 0.7623 on the testing dataset of AAV mutants validated by Bryant et al. (2021), a 14.51% improvement over the CNN trained on the C1R2 dataset reported in Bryant et al. (2021). This project demonstrates proof-of-concept for successfully applying NLP to capsid engineering. PENNSYLVANIA Smart Club: A Sensing Device for Tracking Club Head Speed and Trajectory Sarah Huang State College Area High School, State College, PA Golf is a sport that requires countless hours of practice and considers many different factors, including club speed and trajectory. Having a high club speed indicates that the ball will leave the ground at a high velocity and therefore travel further. In addition, the trajectory which the club travels is important as having the proper swing mechanism can improve the speed of the club dramatically. Therefore, it is imperative that a golfer have access to technology that can detect the club speed and trajectory of the ball in real time. Currently, there are commercial products that provide these, including the Trackman, but it is not portable and therefore cannot be brought onto the course. Therefore, there arose a need for a portable device that would provide the club speed and trajectory in real time. Through deriving an algorithm with the use of concepts such as relative motion and vectors and coding it into MATLAB, a program was devised to provide the club speed and then subsequently graph the 3-dimensional trajectory in real time, called the Smart Club. After preliminary testing, it was found that the results closely model the numbers given by commercial devices like the Trackman. Currently, testing is being continued to further shine light on the accuracy of the proof-of-concept design in comparison to the Trackman.

Competition history

• JSHS 2023 Category not listed

Resources

• National JSHS abstract book ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.