A CNN model for predicting binding affinity changes between SARS-CoV-2 spike RBD variants and ACE2 homologues

Area of Science:

• Virology
• Computational Biology
• Genomics

Background:

• Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) entry relies on the interaction between its receptor binding domain (RBD) and human angiotensin-converting enzyme 2 (hACE2).
• Predicting how amino acid substitutions affect RBD-hACE2 binding affinity is crucial for public health surveillance and understanding viral adaptation in non-human species.

Approach:

• A convolutional neural network (CNN) model was developed, trained on protein sequence and structural features.
• The model predicts experimental RBD-hACE2 binding affinities for 8,440 variants with single and multiple amino acid substitutions.
• The CNN model achieved 83.28% classification accuracy and a 0.85 Pearson correlation coefficient in cross-validation tests.

Key Points:

• The model predicts increased binding affinity for most currently circulating SARS-CoV-2 variants.
• Exhaustive screening identified novel RBD variants with enhanced binding to human and animal ACE2 receptors.
• Binding affinity trends against animal ACE2 receptors mirror those against human ACE2.

Conclusions:

• White-tailed deer ACE2 exhibits strong binding affinity with RBD, comparable to human ACE2.
• Cattle, pig, and chicken ACE2 receptors show weaker binding affinities.
• The model can assess if viral adaptation to animal hosts impacts binding with human ACE2 or reduces overall fitness.