Where can we go from here?

• Improved (simplified?) measures of shape

• Predict shape measure and link this to antigen class

• Skip the shape measure and predict antigen class directly

  • Use CDR length directly - maybe composition?

Notes: