ACD/LogP DB
vs. Competition
|
|
March 2000
"A Rebuttal Regarding Recent Comparisons of SciLogP Ultra (Scivision, Inc.) with ACD/LogP Prediction Software."
Antony Williams and Eduard Kolovanov
Read the article
in PDF format (946 Kb) or MS Word 97 format (2097
Kb zip file).
ACD/Labs continues to test the performance of their prediction tools against published data. We are overjoyed when publications containing comparisons are published, as they offer us the opportunity to test ourselves against the results of the third parties. We have performed such comparisons a number of times, and you can review them at the following web pages:
http://www.acdlabs.com/publish/logp_rev98/
http://www.acdlabs.com/products/phys_chem_lab/logp/competit1.html and
http://www.acdlabs.com/products/phys_chem_lab/train.html
Below we report one of our recent comparisons.
Calculation Procedures for Molecular Lipophilicity:
A Comparative Study
Raimund Mannhold (1) and Karl Dross (2)
(1) Heinrich-Heine-Universitat, Institut für Lasermedizin, Arbeitsgruppe
Molekulare Wirkstoff-Forschung, Universitatsstrase 1, D-40225 Dusseldorf,
Germany
(2) Heinrich-Heine-Universitat, C. und ). Vogt Institut für Hirnforschung,
Universitatsstrase 1, D-40225 Dusseldorf, Germany
The predictive power of 14 calculation procedures for molecular lipophilicity was checked by comparing with reliable experimental LogP values from the literature. The database of 138 test compounds comprised 90 simple organic structure and 48 chemically heterogeneous drug molecules (beta-blockers, class I antiarrhythmics and neuroleptics). The investigation led the authors to confirm that the predictive power of the calculation procedures was significantly better for simple organic molecules than for chemically heterogeneous drug structures. The calculation procedures were arranged in three groups with significantly differing predictive power: fragmental > atom-based > conformation-dependent approaches. The approach utilized within ACD/LogP is a fragment based approach and we are certainly not surprised to see that this was borne out in this study.
The table below reports the data analysis for the 14 programs analyzed PLUS those results reported from our program. As you can see ACD/LogP produces the highest R-factor and the lowest number of disputable values when compared head-to-head with the other programs, an indication of the quality of our predictions as well as the rigorous checking of data used to derive the prediction algorithms.
| |
ACD/Labs LogP DB
4.0 |
f-SYBYL |
SANLOGP
ER |
PROLOGP
cdr |
CLOGP
4.34 |
| acceptable |
95.6 |
81.9 |
79.7 |
76.8 |
84.8 |
| disputable |
4.4 |
13.8 |
15.2 |
16.7 |
10.1 |
| unacceptable |
0.0 |
4.3 |
3.6 |
5.1 |
3.6 |
| not calculat. |
0.0 |
0.0 |
1.4 |
1.4 |
1.4 |
| >logP |
40.1 |
59.4 |
58.7 |
47.8 |
44.2 |
| <logP |
48.2 |
39.1 |
38.4 |
47.8 |
47.1 |
| a ± c.l. |
0.998 ± 0.006 |
1.021 ± 0.023 |
1.031 ± 0.021 |
1.016 ± 0.024 |
1.009 ± 0.021 |
| s |
0.236 |
0.444 |
0.402 |
0.448 |
0.398 |
| r |
0.987 |
0.959 |
0.967 |
0.957 |
0.965 |
| F |
5094 |
1583 |
1919 |
1472 |
1849 |
| |
KLOGP
4.0 |
KOWWIN |
PROLOGP
comb. |
MOLCAD |
Tsar 2.2
4.34 |
| acceptable |
84.1 |
90.6 |
81.2 |
68.1 |
68.1 |
| disputable |
13.8 |
5.8 |
15.2 |
20.3 |
30.3 |
| unacceptable |
0.7 |
3.6 |
2.2 |
11.6 |
11.6 |
| not calculat. |
1.4 |
0.0 |
1.4 |
0.0 |
0.0 |
| >logP |
38.4 |
37.7 |
35.5 |
29.0 |
29.0 |
| <logP |
59.4 |
58.7 |
62.3 |
69.6 |
69.6 |
| a ± c.l. |
0.976 ± 0.019 |
0.984 ± 0.018 |
0.939 ± 0.021 |
0.882 ± 0.023 |
0.877 ± 0.023 |
| s |
0.362 |
0.334 |
0.387 |
0.439 |
0.438 |
| r |
0.966 |
0.974 |
0.960 |
0.932 |
0.937 |
| F |
1859 |
2517 |
1582 |
911 |
987 |
| |
PROLOGP
atom |
CHEMICALC-2 |
SMILOGP
ER |
HINT
cdr |
ASCLOGP
4.34 |
| acceptable |
76.8 |
68.8 |
49.3 |
68.1 |
55.1 |
| disputable |
14.5 |
17.4 |
24.6 |
15.9 |
28.3 |
| unacceptable |
7.2 |
13.8 |
18.8 |
13.8 |
15.2 |
| not calculat. |
1.4 |
0.0 |
7.2 |
2.2 |
1.4 |
| >logP |
31.2 |
21.7 |
10.1 |
40.6 |
48.6 |
| <logP |
65.9 |
74.6 |
81.9 |
52.9 |
49.3 |
| a ± c.l. |
0.911 ± 0.023 |
0.886 ± 0.028 |
0.838 ± 0.033 |
1.016 ± 0.036 |
0.987 ± 0.041 |
| s |
0.431 |
0.535 |
0.588 |
0.682 |
0.771 |
| r |
0.947 |
0.926 |
0.917 |
0.912 |
0.873 |
| F |
1164 |
827 |
660 |
665 |
431 |
Programs and methods for logP calculation, included in the present study
| Calculation Approach |
Method |
Computer Program |
| fragmental methods |
| Rekker, original version |
fragmental constants |
PROLOGP 5.1 cdr |
| Rekker, revised version |
fragmental constants |
SYBYL 6.2 with SPL
macro logp.spl |
| fragmental constants |
SANALOGP ER |
| Leo, Hansch |
fragmental constants |
CLOGP 4.34 |
| Klopman |
computer-identified fragments (CASE) |
KLOGP |
| Meylan, Howard |
atom/fragment contributions |
LOGKOW KOWWIN |
| atom-based approaches |
| Ghose-Crippen |
atomic values |
MOLCAD |
| atomic values |
Tsar 2.2 |
| atomic values |
PROLOGP 5.1_atomics |
| Suzuki |
atomic values |
CHEMICALC-2 |
| Dubost |
atomic contributions |
SMILOGP |
| combined fragmental and atom-based approach |
| Darvas, Csizmadia |
atomic values and fragmental constants |
PROLOGP 5.1_comb |
| conformation-dependant approaches |
| Abraham, Kellog |
|
HINT |
| Ulmschneider |
approximate surface calculation |
ASCLOGP |
|
