New Insights Into Drug Repurposing for COVID-19 Using Deep Learning

Abstract

The coronavirus disease 2019 (COVID-19) has continued to spread worldwide since late 2019. To expedite the process of providing treatment to those who have contracted the disease and to ensure the accessibility of effective drugs, numerous strategies have been implemented to find potential anti-COVID-19 drugs in a short span of time. Motivated by this critical global challenge, in this review, we detail approaches that have been used for drug repurposing for COVID-19 and suggest improvements to the existing deep learning (DL) approach to identify and repurpose drugs to treat this complex disease. By optimizing hyperparameter settings, deploying suitable activation functions, and designing optimization algorithms, the improved DL approach will be able to perform feature extraction from quality big data, turning the traditional DL approach, referred to as a "black box," which generalizes and learns the transmitted data, into a "glass box" that will have the interpretability of its rationale while maintaining a high level of prediction accuracy. When adopted for drug repurposing for COVID-19, this improved approach will create a new generation of DL approaches that can establish a cause and effect relationship as to why the repurposed drugs are suitable for treating COVID-19. Its ability can also be extended to repurpose drugs for other complex diseases, develop appropriate treatment strategies for new diseases, and provide precision medical treatment to patients, thus paving the way to discover new drugs that can potentially be effective for treating COVID-19.

Fig. 1.

Schematic of the structure of SARS-CoV-2 and its entry and replication process within the host cell. It has four structural proteins, S (spike), E (envelope), M (membrane), and N (nucleocapsid) proteins; the N protein holds the RNA genome, while the S, E, and M proteins together create the viral envelope . 1— coronavirus enters the host cells by cleavage of its spike protein (S glycoprotein). 2—binds to ACE2 receptor with its spike protein. ACE2 receptor is primed by the TMPRSS2 protease. 3—coronavirus then fuses into the host membranes. 4—viral single-stranded positive RNA is released for replication of virus RNA. 5—translation of coronavirus polymerase. 6 and 7—transcriptions and replications of RNA occur. 8—translation of coronavirus structural protein. 9—nucleocapsid then combined with S, E, and M proteins. 10—formation of coronavirus is completed. 11—released to infect other cells.

Fig. 2.

Comparison of de novo drug discovery and DR journeys. DR begins with compound identification and the acquisition of an existing drug Phase I clinical studies was not required because the results are already available. This is directly followed by phases II and III clinical trials.

Fig. 3.

General deep learning framework used for DR.

Fig. 4.

Warmer color in A indicates features using “attention” mechanism. The highlighted red line in B indicates the binding site of drug and protein sequence. Black regions (C and D) indicate binding sites marked by the attention mechanism , , , . The above shows that model with attention mechanism can give an insight on biological interpretation .

Fig. 5.

Illustration shows the concept of transfer learning. (a) Source sample consists of huge volume of drug datasets to predict the drug side effects. Random weights were trained to achieve learned weights on source sample. The learned features from (a) were transferred to (b), for example, the weights of the network were transferred to (b). (b) Target sample that includes drugs SMILES and FASTA. The learned weights on the source sample are then trained to achieve fine-tuned weights on source sample.

Fig. 6.

Illustration of four proposed encoder–decoder frameworks for predicting repurposed drug candidates for COVID-19. The two inputs are represented by molecular structures and protein sequences. After training the task-specific datasets using pretrained BERT-based models, the outputs from the pretrained encoders are concatenated and directed to a decoder. (a) Two inputs, drug molecules (SMILES) and protein molecules (FASTA), which are processed by BERT. The outputs are then concatenated before processed using multilayer perceptron with an activated softmax to predict repurposed drug candidates for COVID-19. (b) Two inputs, drug molecules (SMILES) and protein molecules (FASTA). The two inputs are processed by BERT and the outputs are concatenated before processed by transformer with activated softmax to predict repurposed drugs. (c) Two inputs, drug molecules (SMILES) and protein molecules (FASTA). The two inputs are processed by BERT and the outputs are then concatenated before processed using either LSTM or CNN with activated softmax to predict repurposed drug candidates for COVID-19.