Performance with Ala-Ala-Ala-Ala

We are using JChem-3.2.11 Oracle cartridge on Linux and have run into very poor performance when searching with Ala-Ala-Ala-Ala-Ala chain. See search results below:


SQL> select smiles from structure where jc_compare(smiles, 'CNC(C)C(=O)NC(C)C(=O)NC(C)C(=O)NC(C)C(O)=O', 't:e tautomerSearch:y stereoSearch:n') = 1;


no rows selected


Elapsed: 00:00:32.70


SQL> select smiles from structure where jc_compare(smiles, 'CC(N)C(=O)NC(C)C(=O)NC(C)C(=O)NC(C)C(=O)NC(C)C(O)=O', 't:e tautomerSearch:y stereoSearch:n') = 1;


no rows selected


Elapsed: 00:02:28.05


SQL> select smiles from structure where jc_compare(smiles, 'CC(N)C(=O)NC(C)C(=O)NC(C)C(=O)NC(C)C(=O)NC(C)C(=O)NC(C)C(O)=O', 't:e tautomerSearch:y stereoSearch:n') = 1;


no rows selected


Elapsed: 00:04:32.51


SQL> select smiles from structure where jc_compare(smiles, 'CC(N)C(=O)NC(C)C(=O)NC(C)C(=O)NC(C)C(=O)NC(C)C(=O)NC(C)C(=O)NC(C)C(O)=O', 't:e tautomerSearch:y stereoSearch:n') = 1;


no rows selected


Elapsed: 00:04:34.74





Do you have any suggestions to improve search performanc?

Hi Matt,





The slow performance is due to the fact that there are too many tautomers of the given query peptide. There are a few things that can be done to avoid slow search:

• PERFECT search (that we use for duplicate detection) is not very different from EXACT search, but is much faster than that. It can take the search parameters that you used, so "t:p" instead of "t:e" would enhance performance. (It is about three times faster for my table with your query and options.) For the description of these two search types, see this section of the documentation.
  
  
  
• In JChem 5.0 we have just released a new table option for JChem tables. It controls that duplicate filtering (PERFECT search) uses canonical tautomers, disallowing tautomer duplicates at duplicate filtering import. When this option is on, there is no need to switch on tautomer searching. See here.
  
  
  
• Alternatively, a custom standardization with canonical tautomerization action could be used. For example, you could use the attached standardizer configuration. (This, as the above table option avoids the combinatorial explosion, but may not be the best for substructure search.)
  
  
  
• In general, the default fingerprint parameters were optimized for drug-like small molecules. If you store peptides, you should probably increase the fingerprint path length parameter to add more information into the fingerprint, and also increase the fingerprint length to compensate the darkening from this parameter change. See this presentation about tuning the fingerprint parameters.
  
  
  

Best regards,


Szabolcs