ACD/LogP DB
What's New
ACD/LogP DB, v. 6.0, features improved prediction and training capabilities and offers a number of new exciting options:
- Expanded internal database now contains 18,412 compounds due to the addition of the BioByte Star List.
- Enhanced accuracy of prediction as a result of an improvement of the logP calculation algorithm for several classes of compounds and an addition of numerous new fragmental and interaction increments. The logP training dataset was substantially increased due to the addition of BioByte data.
- Similarity Search: in addition to the existing structure and substructure searches, you can now search the database to discover compounds with similar chemical structures.
- New property - Polar Surface Area (PSA) is added to the LogP prediction software
- Enhanced user training: logP prediction engine can now be simultaneously trained with multiple databases to improve the accuracy of calculations for user's datasets and to reduce training time. You can also transfer your logP user database to the logP user fragments file to shorten training time even further.
- Enhanced integration with ACD/ChromManager: experimental chromatographic data can be transferred into an ACD/LogP DB training database, transforming retention times into logP/logD data.
- ACD/LogP DB databases can now be viewed in table format.
- SDfile Import/Export options can be customized.
As part of an ongoing effort to improve the scope and predictive accuracy of ACD/LogP DB, in early 2001 Advanced Chemistry Development, Inc. entered into a strategic partnership with BioByte Corp. Under the terms of the agreement, we were able to use BioByte's Star List database of measured logP values to improve the accuracy and scope our logP prediction.
Additionally, the content of the Star List was added to the Internal Database of LogP DB, with primary literature citation and credit to BioByte.
As a result of this collaboration, we were able to review and add thousands of logP values to the ACD/LogP DB training set. The algorithm for version 6.0 of ACD/LogP DB now uses more information about different chemical classes, additive groups and aromatic/aliphatic interactions:
| |
Ver 1.0-4.0 |
Ver. 4.5 |
Ver. 5.0 |
Ver. 6.0 |
| Additive groups |
531 |
830 |
859 |
1240 |
| Aliphatic Interactions |
652 |
873 |
875 |
1506 |
| Aromatic Interactions |
996 |
1019 |
1589 |
2433 |
| Total |
2179 |
2722 |
3323 |
5179 |
We were able to successfully predict the logP values for the compounds from the BioByte Star List using the earlier versions of ACD/LogP DB. However, the addition of new groups and interactions substantially improved the quality of prediction for version 6.0, as shown below. As reflected by the statistical data, the standard error of logP prediction decreased from 0.351 for version 5.0 to 0.217 for version 6.0.
| ACD/LogP version 5.0 | ACD/LogP version 6.0 |
 |
 |
| SE -Standard Error |
Equation: logPexp = a · logPcalc + b
| version |
a |
b |
N |
r2 |
SE |
|---|
| 5.0 |
0.872 |
0.281 |
10,862 |
0.880 |
0.351 |
| 6.0 |
0.961 |
0.095 |
10,897 |
0.955 |
0.217 |
Error distribution of the logP prediction also substantially improved, as demonstrated below using the same Star List testing set of 10,897 compounds. Over 75% of the predictions were within 0.3 units from the experimental value for predictions made by ACD/LogP DB version 6.0, as compared to 63.8% for version 5.0.
| Difference in LOG units |
Number of compounds |
Percent of calculated |
Number of compounds |
Percent of calculated |
| Version 5.0 |
Version 6.0 |
| All database |
10,897 |
|
10,897 |
|
| Calculated |
10,862 |
|
10,897 |
|
| < 14.0 |
10,862 |
100.00% |
- |
- |
| < 9.0 |
10,861 |
99.99% |
- |
- |
| < 6.0 |
10,850 |
99.89% |
- |
- |
| < 5.0 |
10,846 |
99.85% |
- |
- |
| < 4.0 |
10,842 |
99.82% |
10,897 |
100.00% |
| < 3.0 |
10,832 |
99.72% |
10,895 |
99.98% |
| < 2.0 |
10,718 |
98.67% |
10,875 |
99.80% |
| < 1.5 |
10,534 |
96.98% |
10,827 |
99.36% |
| < 1.0 |
10,002 |
92.08% |
10,596 |
97.24% |
| < 0.7 |
9,263 |
85.28% |
10,131 |
92.97% |
| < 0.5 |
8,342 |
76.80% |
9,480 |
87.00% |
| < 0.3 |
6,826 |
62.84% |
8,215 |
75.39% |
| < 0.2 |
5,638 |
51.91% |
7,069 |
64.87% |
| < 0.1 |
3,928 |
36.16% |
5,163 |
47.38% |
| Not calculated |
35 |
|
0 |
|
Database
The database has increased in total number of entries, by 23%. It now contains 14606 structures with one or more literature references and experimentally measured values.
General Interface
The system can now be trained without opening database, the user can just specify the location of the previously created database through the Results Window:
Internal and user-defined databases can be examined now with the Multiple Record View. A right-click gives you a pop-up menu that lets you select display options for zoom, data fields displayed and font.
You can upgrade the indices of database records for more rapid substructure search.
Algorithm
We have recalculated dozens of old, not very reliable, increments.
An improved algorithm for the prediction of LogP uses more information about additive groups and aromatic interactions.
| |
Ver 1.0-4.0 |
Ver. 4.5 |
Ver. 5.0 |
| Additive groups |
531 |
830 |
859 |
| Aliphatic Interactions |
652 |
873 |
875 |
| Total |
2179 |
2722 |
3323 |
Available Statistics for Version 1.0-4.0
Training Set, N = 3688
Standard Deviation (StD) = 0.21
R = 0.99
Equation is unavailable
Available Statistics for Version 4.5
Training Set, N = 4540
StD = 0.2345
R = 0.9893
LogPexp = 0.9753*LogPpred + 0.0586
Available Statistics for Version 5.0
Training Set, N = 5214
StD = 0.2392
R = 0.9882
LogPexp = 0.9757*LogPpred + 0.0569
|
