Table 3. Top-Performing Models for DMSO Solubility Prediction Developed with Enamine Data Set within the Whole Data Set and 10% Coverage^a.

method	descriptors	number of descriptorsb	BAC100%	PPV100%	ENR100%	PPV10%	ENR10%	BAC10%	SENS10%	SPEC10%	NSOL10%
J48	CDK	175	71.8	97.5	2.2	99.4	8.8	87.1	97.8	76.4	2.3
ASNN	E-state	205	73.1	97.9	2.6	99.4	8.8	92	96	88	4.7
ASNN	Dragon6	1929	73.2	97.9	2.5	99.4	8.8	92.2	97	87.4	3.7
ASNN	Fragmentor	872	73.3	97.9	2.5	99.3	7.6	90.2	95.8	84.6	4.3
ASNN	ChemAxon	134	72.8	97.9	2.5	99.3	7.6	91.2	92.4	89.9	5.8
J48	E-state	205	71	97.3	2	99.2	7.6	77.5	99.4	55.6	1.7
ASNN	CDK	175	73.2	97.9	2.5	99.2	6.6	91	94.8	87.2	5.3
J48	Dragon6	1929	71	97.3	2	99.2	6.6	88.8	97	80.6	4
J48	Fragmentor	872	71	97.3	2	99.1	5.9	82.2	98	66.4	2.5
ASNN	Mera	259	70.9	97.7	2.3	99.1	5.9	88.3	96.2	80.4	4.2
ASNN	Adriana	119	72.9	97.9	2.5	99	5.3	86	82.2	89.8	7.8
J48	Adriana	119	69	97.1	1.8	99	5.3	79.4	98.8	60	2.5
RF	Fragmentor	872	72	97.6	2.2	99	5.3	88.3	97.4	79.1	4.7
ASNN	GSFrag	314	71.9	97.8	2.4	99	5.3	87.6	85.5	89.7	7.5
J48	GSFrag	314	69	97.1	1.8	99	5.3	83	98.8	67.1	2.9
RF	E-state	205	73	97.8	2.4	99	5.3	88.5	96.3	80.6	5.1
RF	inductive	38	67	97.2	1.9	99	5.3	84.9	94.8	74.9	3.5
consensus	all		74.2	97.9	2.5	99.6	13	87.5	98	77	1.2

^aModels with PPV_10% ≥ 99% were selected.

^bAfter filtering, see Methods section. NSOL_10% is the percentage of nonsoluble molecules within 10% coverage of the most highly accurate predictions.