Problems searching products by SMILES or SMARTS

[bdeadman]

As reported by a user, and observed by me in #122, the chemical searcher is not finding the expected results.

• Searching for "[C,c]=" in Products, using SMARTS does not find anything on either the production or staging version of interface.
• Searching for "[C,c]#" in Products, using SMARTS does not find anything on either the production or staging version of interface.
• Searching for an exact SMILES of a known product (e.g. CC=C1C=C(OC)N=C(N)N1   ) does not find anything on production or staging version.
• Same as above for searching by similarity or exact structure

[bdeadman]

Running the following on my (out of date) clone of ord-data, I get 411 records.

SELECT compound.smiles AS smiles, compound.reaction_role FROM ord.compound WHERE smiles LIKE 'OCCS' ;

Running the search in the online interface returns no results for 'Reactants & Reagents' = "OCCS" with the exact, similarity or substructure search options.

The same search with the SMARTS option returned 100 entries (limited by the query) but note that when replicating the query it now returns no results.

[bdeadman]

Running the following on my (out of date) clone of ord-data, I get 181 records.

SELECT compound.smiles AS smiles, compound.reaction_role FROM ord.compound WHERE smiles LIKE 'O=C1C=CC(=O)C=C1' ;

Running the search in the online interface returns no results for 'Reactants & Reagents' = "O=C1C=CC(=O)C=C1" with the exact, similarity or substructure search options. SMARTS search also failed to return results. This was on the production and staging instances of ord.

[bdeadman]

@skearnes @miori-nd who wants this one?

[miori-nd]

I'll look into this

Get Outlook for Androidhttps://aka.ms/AAb9ysg

---

From: Ben Deadman @.> Sent: Tuesday, August 6, 2024 1:12:12 PM To: open-reaction-database/ord-interface @.> Cc: miori @.>; Mention @.> Subject: Re: [open-reaction-database/ord-interface] Problems searching products by SMILES or SMARTS (Issue #125)

@skearneshttps://www.google.com/url?q=https://github.com/skearnes&source=gmail-imap&ust=1723569134000000&usg=AOvVaw3DIRgGStY81yK4rZCM4pCL @miori-ndhttps://www.google.com/url?q=https://github.com/miori-nd&source=gmail-imap&ust=1723569134000000&usg=AOvVaw2wx0aNSCsUSEgtl_FrZwJJ who wants this one?

— Reply to this email directly, view it on GitHubhttps://www.google.com/url?q=https://github.com/open-reaction-database/ord-interface/issues/125%23issuecomment-2271760232&source=gmail-imap&ust=1723569134000000&usg=AOvVaw2GFvCsYhccbPDNXi-sU4a-, or unsubscribehttps://www.google.com/url?q=https://github.com/notifications/unsubscribe-auth/BFLMMVRW2CPPXESLKVZ45CTZQD7WZAVCNFSM6AAAAABMCWPQCWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDENZRG43DAMRTGI&source=gmail-imap&ust=1723569134000000&usg=AOvVaw0MfPVe9bSVSWqlAw-vD0Me. You are receiving this because you were mentioned.Message ID: @.***>

[skearnes]

I looked at the product SMILES search; this is a timeout on the backend. I'll dig into the SQL query and see if I can optimize it.

[skearnes]

I can definitely speed up the "exact" queries. Will push up soon.

[skearnes]

@bdeadman I pushed the exact query fix to prod. The SMARTS patterns listed are invalid; put them in https://smarts.plus/smartsview for testing.

[bdeadman]

Thanks @skearnes. I was working of this resource https://www.daylight.com/dayhtml/doc/theory/theory.smarts.html. For my future reference the correct SMARTS pattern would be "[c,C]=[c,C]" and this validates in the tester you linked. The problem was I hadn't ended the SMARTS with a node (an atom).

I'll run some tests on the prod tomorrow.

[bdeadman]

Testing on production interface today. Values in table represent the number of reactions returned. Results limited to 100.

Search Term	Exact	Similar(0.5)	Substructure	SMARTS
"Reactants & Reagents" =
"SCCO"	100	0	0	0
"OCCS"	100	0	0	0
"O=C1C=CC(=O)C=C1"	100	0	0	0
"C1=CC(=O)C=CC1=O"	100	0	0	0
"c1ccncc1"	0	0	100	100
"[c,C]=[c,C]"	NA	NA	NA	100
"[c,C]#[c,C]"	NA	NA	NA	0
"C#C"	100	0	0	100
"Products" =
"SCCO"	0	0	0	0
"OCCS"	0	0	0	0
"CC=C1C=C(OC)N=C(N)N1"	1	0	0	0
"c1ccncc1"	7	0	100	100
"[c,C]=[c,C]"	NA	NA	NA	100
"[c,C]#[c,C]"	NA	NA	NA	0
"C#C"	6	0	100	100

[bdeadman]

Outcomes from above testing:

• Exact searches are working and giving results that seem reasonable.
• The exact search is parsing the SMILES as SMILES and not just a string. The reversed "OCCS" was also found even though this string is not a component in the database.
• Substructure searching only works when small fragments given. I was expecting the substructure search to also output the exact search results.
• I also expected "SCCO" substructure search to find reactants and products like in the image in this comment.
• Similarity searching - same as for substructure searching. I expected it to at least output the exact search results.
• I suspect the above problems with substructure and similarity searching could be a timeout issue. See #34, although it does not appear as a 502 bad gateway error.