Summary of PDBj-DDBJ-KEGG workflow demo

Who we are

•  DDBJ (Sugawara & Shigemoto)
•  PDBj (Standley & Kinjo)
•  KEGG (Kawashima)

Our Task

Somehow integrate DDBJ, PDBj, and KEGG.

After a lot of proposes were rejected, we decided to make a workflow.

What it is

Annotate a protein sequence by homology based on sequence and structure.

Input: Given an unannotated protein sequence ("hypothetical", "putative", etc.)

1. Run BLAST against DAD at DDBJ.
2. When annotated homologs are found, get their annotations from DDBJ.
3. When only hypothetical proteins are found, run BLAST against PDB.
4. If any homologs are found (annotated or not), they are sent to Structural-Navigator (structure search) at PDBj.
5. If any similar structures are found, get the annotations from PDBj and KEGG.

Output: homology-based annotation from DDBJ, KEGG, and PDBj

How we implemented it

• We used Taverna. (Thanks you, Stuart!)
• DDBJ and KEGG have plenty of SOAP services which are actually used.
• PDBj's SOAP services are very poor (no SOAP service for Structure-Navi), but there are some REST services which were used after some modifications.
• Only one of us knew Java (Mr. Shigemoto from DDBJ) so he did most of the programming (BeanShell scripts).

What we learned

• No conditional branch in Taverna.
• We needed to write many small glue codes to connect different services.
• BeanShell scripts are handy.
• But, writing BeanShell scripts requires a Java programmer (We had only Shigemoto-san...).
• SOAP is usable (AK's comment).
• It's nice to have knowledgeable people around. We can ask questions and modify codes instantly.

What can be improved

• Server-side programming?
• Lots of small widgets?
• Knowing the target users. (Who are they, anyway?)

Sort of conclusions…

• It's unrealistic to make strictly standardized data format and/or API.
• First of all, individual providers should hack to make usable services.
• Only after then, we can make interfaces interoperable by hacking.
• Thus, it's nice to have a Hackathon once in a while...