Exploring the Performance of Convolutional Neural Networks for mRNA cancer vaccine 5’ UTR Design

A. Yan
Mercer County Community College,
United States

Keywords: mRNA, immunotherapy, vaccine, machine learning, sequence design, transformer, neural networks


Immunotherapy is a powerful cancer treatment modality that induces or enhances immune responses to specifically target cancer cells. This specificity avoids harming healthy non-tumor cells and limits the side effects related to other primary cancer treatment modalities such as chemotherapy and radiation therapy. Among the many agents of delivery for immunotherapy, mRNA vaccines are notable for their ease of manufacturing and safe administration. They have shown promise in both preclinical studies and clinical trials and warrant further investigation. Many of the studies on mRNA vaccines have focused on improving mRNA stability, decreasing immunogenicity, and improving delivery efficiency, but there is ample room for improvement of the primary sequence of mRNA to increase translation efficiency. As 5’ untranslated regions (UTRs) play a large role in regulating protein expression, they are a prime candidate for optimization. Optimus 5-Prime is a machine learning model based on a convolutional neural network that can predict mean ribosome load (MRL) from a given 5’ UTR sequence alone. This allows us to tune 5’ UTR sequences to give the desired level of protein expression. We plan to improve on Optimus 5-Prime and build a model that gives more accurate MRL prediction and allows for more efficient 5’ UTR sequence engineering.