The entire world is engulfed into a corona pandemic attack. At present, there are 191127 positive cases of noble COVID-19 infection all over the world with total fatalities of 7807 according to a report by the World Health Organization(WHO).
Amidst all the attempts to fight this deadly disease which has forced lockdowns. These lockdowns have led to business losses and work from home culture among several businesses. Governments are trying hard to fight this odd, while the virus seems to have a ball!
While the US began a vaccine trial based on RNA binding proteins with volunteers given a Corona-like virus, but, not exactly the lethal virus for human trials. Though human trials have been initiated rapidly curtailing the animal trials, it can take about 18 months’ time to get an approved vaccine for Corona.
The question here is,
Can we speed up the vaccine creation process? And How?
The answer lies in the technological domain and not the medical domain. Deep Learning is a method based on Artificial Intelligence that allows the usage of backpropagation algorithms to allow AI machines to change its internal metrics to make changes in the computation of every new layer of data representations from the previous layers.
Deep Learning has helped innovative technologies like speech recognition, image detection, facial recognition, and even drug discovery. So, let’s discover how can we use this technology to speed up the process of finding an effective vaccine for COVID-19?
First! What is Deep Learning?
We already discussed how deep learning is concerned with the backpropagation method of algorithm learning. To make this simple, let us compare it with Machine Learning.
Machine learning has three major learning methods:
- Supervised Learning
- Unsupervised learning
- Reinforced Learning
But, all these learning methods rely on structured data, while deep learning leverages layers of ANNs(Artificial Neural Networks). In simple words, Machine learning uses specific labels given to an object for classification and deep learning will use network layers to define hierarchy of features of an object to classify objects.
Now, What are RBPs?
RBP or Reticulocyte Binding Proteins are used as an alternative to viruses and target disease bacterias to evoke human antibodies for protection against diseases. To commemorate the effective genome research and vaccine building, RBP sites are predicted to target the RBP attack like it was used in the case of Plasmodium Vivax Malaria in Africa.
RBP target sites are predicted based on RNA-interactions and there has been comprehensive research on the same with many existing models in place. Two of the most popular methods used are:
Feature Level Fusion Method:
Predictors of RNA-interaction sites are developed through sequence information. We all know that everything about the human genome and disease research these days needs RNA-sequencing. It is like a coded language we use in an app development structure.
These sequencing data is collected from different sources across cellular activities and changes in genome sequences. The fusion of these data and features into a single high-dimensional feature is used to predict RBP. But, it has dimensional drawbacks and needs more time.
Decision-Level Fusion:
In this method, five different learning models are proposed. These models work on different aspects of RNA-interactions like Genome sequence. Secondary sequence, Genome Ontology information, region type for RNA-interaction, and others. Results from these models are fused in the end to predict an RNA-interaction site.
Using A New Deep Learning Approach for RBP site predictions:
Discovering RNA-interaction sites or RBP sites is essential for genome-based research, especially in the field of drug discovery. Here, we are exploring a deep learning method called iDeep for predictions of RBP sites to enhance the vaccine discovery of novel Coronavirus.
We already saw two predictive models popularly used by researchers, wherein there lies a common drawback. Both the models used features extracted from observed data, which can be subject to error.
Deep Learning provides a unique approach, which works with hybrid multiple abstraction layers. These abstraction layers are used to amp data with a high-level abstraction space. The unique thing about Deep Learning-based models is that they integrate heterogeneous data and learn complex patterns from raw input.
Deep Learning Framework for RBP site prediction in vaccine research for Corona:
CNN(Convolutional Neural Network):
It is a Deep Learning model that is different from traditional statistical learning models. It combines feature extraction and learning of patterns or features in a single step rather than doing in two different steps. This reduces the possibility of a mismatch between features extracted and features learned by the model.
CNN models are used for recognizing RNA-motifs by identifying the patterns based on filters employed by the model on the input data. RNA-motifs are a special RNA sequence used for the construction of an RNA structure.
DBN(Deep Belief Network):
It is another Deep Learning algorithm that optimizes the learning of high-level features from a huge amount of data. Deep-net rbp is an algorithm used to predict RBP interaction sites with DBN.
iDeep Framework:
iDeep is a multi-model framework created through a combination of several CNN and DBN models. This hybrid network uses CNN models for sequential data and DBN models for binary or numeric data. Different deep neural networks combined in a hybrid framework will be trained using raw input data.
Further, results from training of these models are tuned from different abstraction layers and a backpropagation approach is used for tuning of abstraction layers from top common layer shared by each model to bottom individual abstraction layers.
Next is the extraction of latent features across the models which are further used in learning of deep learning algorithms to predict RBP sites for targeting the RNA-interactions.
Now that you have learned about a new approach to finding the Corona vaccine, let’s discover how exactly it can help researchers achieve it faster?
- Corona vaccine discovery is now into human trials and there is a need to identify RBP sites to target the mRNAs effectively for better results.
- iDeep explores the multi-layered hybrid framework that allows learning of models layer-by-layer.
- By this approach, researchers can enhance the detection of RNA-interaction sites through successive layer learning of features.
- iDeep model helps with the creation of a process where the output of a layer acts as the input of a successive layer.
- It integrates CNN and DBN, which individually helps the process. CNN helps learn and capture regulatory motifs of RNA sequencing.
- While DBN can capture and extract high-end levels of features from the raw input data.
- Thus, enhancing the classification abilities by tapping into different sources of RNA-protein binding.
- The fusion of a shared abstraction layer along with bottom individual layers makes the handling of the features quite easy for the iDeep framework.
- In comparison with current existing frameworks, iDeep can easily pace-up RNA- interaction site prediction and render error-free results.
Conclusion:
The new age plague has hit our world with brutal blow and Artificial Intelligence can be that modern frontier that can help our fightback. Though AI has been disruptive in the fields of mobile app development and its revolutionary applications have been phenomenal.
But, if researchers can leverage the deep learning approaches to pace-up the corona vaccine discovery, then many lives can be saved and the world will be back in business.