Table 2 Number of PE09 samples (out of 7) correctly recognized as outliers for the CA04 antigenic cluster. First row: samples predicted to be outside a 95% C.I. Centered on the cluster centroid. Second row: samples predicted to be outside a 2 A.u. Radius Centered on the cluster centroid.

From: Language models learn to represent antigenic properties of human influenza A(H3) virus

# of PE09 samples recognized as outliers (out of 7)

Genetic distance

Physicochemical signature

BiLSTM

ProtBERT

95% C.I.

2

4

7

6

2 a.u. radius

1

3

5

4