Skip to main content

Table 3 The comparison of the performance of the Morgan2 (ECFP4), rv-QAFFP and b-QAFFP fingerprints for biological activity classification of 23 CLASS data sets

From: QSAR-derived affinity fingerprints (part 1): fingerprint construction and modeling performance for similarity searching, bioactivity classification and scaffold hopping

FP

Morgan2

rv-QAFFP

b-QAFFP

AD

–

No

Yes

No

Yes

Cutoff

–

–

–

5

6

7

8

5

6

7

8

AUC

0.87 ± 0.01

0.86 ± 0.01

0.86 ± 0.02

0.83 ± 0.01

0.85 ± 0.01

0.84 ± 0.02

0.77 ± 0.01

0.85 ± 0.02

0.85 ± 0.02

0.83 ± 0.02

0.73 ± 0.01

EF5

2.16 ± 0.16

2.08 ± 0.14

2.08 ± 0.13

2.03 ± 0.13

2.09 ± 0.14

2.08 ± 0.14

1.89 ± 0.11

2.10 ± 0.14

2.08 ± 0.14

2.04 ± 0.13

1.78 ± 0.10

  1. Model AD was estimated by an ICP with the confidence level of 90%. rv-QAFFP models were trained using raw data. Considering AD for rv-QAFFP means that if the prediction interval width was larger than ± 2.0, the prediction was regarded unreliable and was replaced by the average of all reliably predicted affinities. Various affinity cutoffs were used to construct the b-QAFFP fingerprint. Affinities predicted to lie outside model AD were encoded by zeros. Best results are shown in columns in italic. Data shown are averages over all CLASS data sets with their standard errors of the mean. Both rv-QAFFP and b-QAFFP fingerprints are 440 bits long