Machine learning guided design of high affinity ACE2 decoys for SARS-CoV-2 neutralization

Area of Science:

• Biochemistry
• Virology
• Computational Biology

Background:

• Engineering high-affinity soluble ACE2 decoy proteins is a strategy to neutralize SARS-CoV-2 by blocking viral spike protein binding.
• A previously identified triple mutant ACE2 variant (ACE2₂.v.2.4) showed nanomolar affinity for the SARS-CoV-2 RBD domain.

Approach:

• A transfer learning algorithm (TLmutation) was employed to identify ACE2 double mutants with high binding affinity to the SARS-CoV-2 RBD.
• The TLmutation model was trained on single mutation effects to predict double mutant performance.
• Focus was placed on identifying variants with reduced mutational load compared to existing high-affinity variants.

Key Points:

• Several ACE2 double mutants demonstrated enhanced binding affinity to the RBD compared to wild-type ACE2.
• The L79V;N90D double mutant exhibited binding affinity comparable to the ACE2₂.v.2.4 triple mutant.
• Experimental validation confirmed the efficacy of the identified double mutants.

Conclusions:

• Transfer and supervised learning are effective for engineering protein-protein interactions.
• High-affinity ACE2 peptides can be identified using computational approaches for targeting SARS-CoV-2.
• This work provides promising therapeutic candidates for combating SARS-CoV-2 infection.