Investigates a blinded, prospective benchmark where forty-nine in silico antibody design submissions were evaluated against experimental affinity and developability data to test computational prediction performance.
Key Takeaways
- Amino acid sequence is not a functionally readable ‘language’
- Exceptions are the heavy and light third complementarity-determining regions (CDR3s)
- Unlike natural language, the statistically dominant token is often not optimal for affinity improvement 11