Synergizing Attribute-Guided Latent Space Exploration (AGLSE) with Classical Molecular Simulations to Design Potent Pep-Magnet Peptide Inhibitors to Abrogate SARS-CoV-2 Host Cell Entry

Abstract

The COVID-19 infection, caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), has evoked a worldwide pandemic. Even though vaccines have been developed on an enormous scale, but due to regular mutations in the viral gene and the emergence of new strains could pose a more significant problem for the population. Therefore, new treatments are always necessary to combat future pandemics. Utilizing an antiviral peptide as a model biomolecule, we trained a generative deep learning algorithm on a database of known antiviral peptides to design novel peptide sequences with antiviral activity. Using artificial intelligence (AI), specifically variational autoencoders (VAE) and Wasserstein autoencoders (WAE), we were able to generate a latent space plot that can be surveyed for peptides with known properties and interpolated across a predictive vector between two defined points to identify novel peptides that exhibit dose-responsive antiviral activity. Two hundred peptide sequences were generated from the trained latent space and the top peptides were subjected to a molecular docking study. The docking analysis revealed that the top four peptides (MSK-1, MSK-2, MSK-3, and MSK-4) exhibited the strongest binding affinity, with docking scores of -106.4, -126.2, -125.7, and -127.8, respectively. Molecular dynamics simulations lasting 500 ns were performed to assess their stability and binding interactions. Further analyses, including MMGBSA, RMSD, RMSF, and hydrogen bond analysis, confirmed the stability and strong binding interactions of the peptide-protein complexes, suggesting that MSK-4 is a promising therapeutic agent for further development. We believe that the peptides generated through AI and MD simulations in the current study could be potential inhibitors in natural systems that can be utilized in designing therapeutic strategies against SARS-CoV-2.

Keywords: Omicron variant; SARS-CoV-2; Wasserstein autoencoders (WAE); deep learning; molecular docking; molecular dynamics simulation; variational autoencoders (VAE).

Figure 1

The model stages of antiviral-pep generation with attribute guidance. (A) Applying amino acid mutations based on affinity, stability, and toxicity to protein structures from PDB to create peptide datasets. (B) Training VAE/WAE model to map peptide sequences into latent space with attribute-guided exploration. (B*) A simplified abstract of the VAE/WAE model. (C) Sampling the latent space to generate peptides with specific attributes, followed by structure prediction using AlphaFold 3.0 and molecular dynamics simulations.

Figure 2

(A) The binding poses and LigPlot analysis of RBD and human ACE2 protein (B). Molecular docking poses of the constructed peptides with RBD.

Figure 3

Docking analysis. (A) MSK-1 with RBD (B) MSK-2 with RBD Omicron variant.

Figure 4

Molecular docking analysis of (A) MSK-3 and (B) MSK-4 with RBD.

Figure 5

RMSD of four peptide complexes. (A) MSK-1 (B) MSK-2 (C) MSK-3, and (D) MSK-4. The y axis represents RMSD Å while the x axis represents time in ns.

Figure 6

Root mean square fluctuation analysis of four complexes (A) MSK-1, (B) MSK-2, (C) MSK-3 and (D) MSK-4.

Figure 7

Radius of gyration of four complexes (A) Rg of MSK-1 complex, (B) MSK-2, (C) MSK-3, and (D) MSK-4.

Figure 8

Solvent accessible surface area of four peptides in complex with Omicron RBD. (A) MSK-1 (B) MSK-2 (C) MSK-3, and (D) MSK-4.

Figure 9

Hydrogen bond analysis (A) MSK-1 complex, (B) MSK-2, (C) MSK-3, and (D) MSK-4.

Figure 10

Dynamic cross-correlation of four peptides with RBD Omicron variant. (A) MSK-1 complex, (B) MSK-2 (C) MSK-3, and (D) MSK-4.

Figure 11

Principal components analysis of four peptides with RBD (A) MSK-1 (B) MSK-2, (C) MSK-3, and (D) MSK-4.

Figure 12

MMGBSA analysis of the top four peptides in complex with SARS-CoV-2.