Table 1.
Formulas for similarity/dissimilarity coefficients for binary-valued vectors.
Fig 1.
Architecture of the WS-ELM.
Table 2.
The 17 activity classes in the MUV dataset.
The entries are ranked in decreasing order of average mean pairwise similarity across four fingerprints.
Table 3.
Maximum percentage actives retrieved in top 1% of ranked database using similarity searching technique (average across 10 runs).
Bold face is the best result in each activity class.
Table 4.
Maximum percentage actives retrieved in top 1% of ranked database using WS-ELM technique (average across 10 runs).
Bold face is the best result in each activity class.
Table 5.
Ranks assigned to 16 similarity coefficients–similarity searching–by 17 activity classes from Table 3.
Table 6.
Ranks assigned to 16 similarity coefficients–ELM–by 17 activity classes from Table 4.
Fig 2.
Relative improvement/worsening with respect to similarity searching for top 1% retrieved–average across ten runs, 16 similarity coefficients, and four fingerprints.
Fig 3.
Violin plot of maximum percentage of active molecules retrieved in the top 1% with WS-ELM in conjunction with 16 different similarity coefficients–averaged across ten runs, 17 activity classes, and four fingerprints.
Fig 4.
Maximum percentage of active molecules retrieved in the top 1% with WS-ELM and similarity searching in 17 activity classes–averaged across ten runs, 16 similarity coefficients, and four fingerprints.
Fig 5.
Maximum percentage of active molecules retrieved with WS-ELM and similarity searching using four different fingerprints–averaged across 17 activity classes, 16 similarity coefficients, and 10 runs.
Table 7.
The percentage hit rate in the top 1% of the ranked database retrieved by WS-ELM and CWS-ELM in conjunction with Jaccard/Tanimoto (JT) and Sokal/Sneath(1) (SN1).
Figures in bold face represent the best performance.
Table 8.
Ranks assigned to the performances of 6 classifiers by 17 activity classes from Table 7.
Fig 6.
Effect of AUROC when the number of hidden nodes in WS-ELM and CWS-ELMKMC is changed in activity class I01.
Solid lines represent mean values while shaded areas represent error/confidence bounds. The upper and lower bounds of each node are based on the standard deviation.
Fig 7.
Effect of AUROC when the number of hidden nodes in WS-ELM and CWS-ELMKMC is changed in activity class I17.
Solid lines represent mean values while shaded areas represent error/confidence bounds. The upper and lower bounds of each node are based on the standard deviation.
Fig 8.
Enrichment plot for the top 1% of the sorted library for each performer with ECFP_6 fingerprint on activity class I01.
Fig 9.
Enrichment plot for the top 1% of the sorted library for each performer with ECFP_6 fingerprint on activity class I17.
Fig 10.
Molecules retrieved by different methods in top 1% of the ranked database for activity class I01.
Fig 11.
Molecules retrieved by different methods in top 1% of the ranked database for activity class I17.
Fig 12.
Early recognition criteria suggested by [35, 38].
(Left) EF (Right) Ratio of true positive rate to the false positive rate, at 0.5%, 1.0%, 2.0%, and 5.0% of the ranked database for WS-ELM and its variants, SVM, RF, and Similarity Searching (SS). Each bar represents the mean value across all activity classes and ten runs.
Fig 13.
Bar charts showing mean EF and BEDROC at 1.0% of the ranked database for WS-ELM and its variants, SVM, RF, and Similarity Seaching (SS).
According to Truchon & Bayly, the top 1% of the ranked database is equivalent to α = 160.9 of BEDROC [34].