The Need for NanoGen:
mRNA vaccines have taken the lead as the fastest vaccine candidates in accelerated treatment development for widespread epidemics. In this era of COVID-19, vaccine discovery is of utmost importance to win the fight against the pandemic that infected millions. However, there is a major setback with these vaccines currently in clinical trials: the need for intense refrigeration or freezing for storage.
This storage technique is not only unsustainable and expensive, it is also a barrier to global distribution. With the risk of vaccine degradation in hot climates, the vaccines that obtain FDA approval will only reach a tiny fraction of the population.
Introducing NanoGen Predict:
We provide a service called NanoGen Predict to pharmaceutical companies in this field of research. NanoGen is an mRNA vaccine sequence AI that predicts degradation reactivity probabilities at every base pair under differing conditions. As one cut in the mRNA sequence will render the whole vaccine useless, we hope to expedite mRNA stabilization research and vaccine discovery by identifying mRNA degradation-prone sites with our AI technology. NanoGen takes the next leap forward in making far-reaching and sustainable mRNA vaccine distribution possible.
How NanoGen works:
NanoGen is based on a Recurrent Neural Network (RNN) model and is trained on the Eterna RNA sequence dataset, a dataset comprised of over 3000 sequences. After processing, NanoGen generates a reactivity probability at every base pair in the sequence under 50 degrees C with and without the treatment of Mg.
What's next for NanoGen:
We have a plan to partner with pharmaceutical companies to aid in their mRNA research. We plan to expand NanoGen Predict to be compatible with large sequence files, so it can analyze more data in one file submission, further expediting the research process. We also plan on connecting NanoGen to a public sequence database so that thousands of sequences can be accessed and analyzed.
I had absolutely no idea what I was doing. This was my first time working with GRU neural networks. I also had trouble integrating the model into the website. Time was an issue as well - I needed to manage my time well enough to allow time for both model construction and training (took around 6 hours for 75 epochs).