Article Preview
Top1. Introduction And Problematic
During the last two decades, the development of prophylactic vaccines (based on mRNA) has attracted increasing interest. Messenger RNA (mRNA) or messenger ribonucleic acid is a molecule found in all cells where a system reads the information encoded in mRNA and converts it into protein (as shown in figure 2). The mRNAs contained in the COVID-19 PFIZER vaccine are converted into S protein (Spike) on the envelope of the Sars-Cov-2 virus. During an infection, this protein penetrate to the cell and trigger an immune response. When a vaccinated person is infected with SARS-Cov-2, his immune system recognizes the S protein of the virus’s and destroys it (Patel, 2020).
The mRNA vaccines activate both parts of the immune response (NRM (non-replicating messenger) and SAM (self-amplifying messenger)) sufficiently intense to dispense with adjuvants as shown in the figure 1. The Pfizer’s vaccine has proved his immunogenic capacity, but we have no details concerning the nature, duration and variations of the immune response induced by its mRNAs (Pashchenko, 2021).
Figure 1.
The fate of the two types of constructions in an RNA vaccine (NRM (non-replicating messenger) and SAM (self-amplifying messenger)) are staggered in cell cytoplasm. The antigens generated are presented to the cells of the immune system via MHC. They are also secreted by cells and are carried in the bloodstream. The mRNA does not alter the cells' genomes or produce transgenic cells. (Patel, 2020).
Unfortunately, one of the major challenges in creating these vaccines is the fragility of the RNA molecule; they can degrade quickly in a few minutes to a few hours and therefore need to be freeze-dried or incubated at low temperatures to remain stable. Pfizer announced that his vaccine should be maintained in -80°C. However, it is still unclear exactly which parts of RNA molecules are most susceptible to spontaneous degradation and therefore difficult to predict the reactivity and degradation of mRNA. The RNA molecules tend to degrade spontaneously which is a serious limitation - a single cut can render the mRNA vaccine useless.
Figure 2.
Depiction of mRNA vaccine-induced antibody response against SARS-CoV-2 spike proteins (Patel, 2020).
In this article, our aim is to introduce three new configuration models of deep learning for this issue using the Stanford OpenVaccine dataset. These models are variants of recurrent neural networks (RNNs) :
- 1)
Model 1 long-term memory networks (LSTMs),
- 2)
Model 2 Bidirectional long-term memory networks (BDLSTMs).
- 3)
Model 3 Bidirectional Encoder Representations from Transformers (BERT).
The purpose is to predict the reactivity and degradation of mRNA molecules with the degree of degradation at each position as depicted in the figure 1. To analyse each part of mRNA, we have used the n-gram codon technique for tokinezation and TF-IDF for vectorization of each RNA structure.
Figure 3.
An example of safed and degradated position of mRNA modecule (Jackson, 2020)