search

trvrb / flux

Integrating influenza antigenic dynamics with molecular evolution
http://bedford.io/papers/bedford-flux/
12 stars 5 forks source link

Performance of purely genetic model #30

Closed trvrb closed 11 years ago

trvrb commented 11 years ago

The authors should add a column to Table 1 showing the performance of the best possible model arising from purely genetic data -- that is, 2-D or 3-D embeddings based on aa hamming distances alone, ignoring all HI data.

trvrb commented 11 years ago

We agree that it would be ideal to provide a test error associated with a purely genetic model. However, test errors in Table 1 are based on the prediction of a set of HI titers based on other HI titers or other HI titers plus genetic data. Without including some form of HI information, it's difficult to see how to predict HI titer. One could have a 'distance' in terms of AA rather than antigenic map units, but in this case, one would still need to include a scaling for 'serum effects', which could not be made without reference to HI data. Regardless, we've attempted to address this issue in our discussion of correlation between antigenic and genetic distances, as described above.