Skip to main content
. Author manuscript; available in PMC: 2021 Jul 14.
Published in final edited form as: J Chem Inf Model. 2020 Aug 4;60(8):3722–3730. doi: 10.1021/acs.jcim.0c00263

Table 1.

TransFS causes CNNs trained on the DUD-E dataset to generalise better to an external validation set than OriginalFS. PR-AUC values for the 14 targets in the ChEMBL validation set for OriginalFS, TransFS and RedockedFS CNNs. Highest scores are highlighted in bold.

Chembl Target Family OriginalFS TransFS RedockedFS
10752 Kinase 0.289 0.272 0.270
12670 Kinase 0.363 0.374 0.350
20014 Kinase 0.573 0.580 0.566
10378 Protease 0.047 0.052 0.032
10498 Protease 0.056 0.078 0.073
11534 Protease 0.062 0.095 0.100
219 GPCR 0.116 0.153 0.069
11279 GPCR 0.039 0.039 0.040
11631 GPCR 0.245 0.264 0.191
12968 GPCR 0.009 0.011 0.011
28 Other (Thymidylate synthase) 0.286 0.363 0.297
276 Other (Phosphodiesterase 4A) 0.318 0.415 0.269
11359 Other (Phosphodiesterase 4A) 0.214 0.270 0.133
18061 Other (Sodium channel protein type IX α-subunit) 0.031 0.031 0.032