Quality label
Q(experts) = (100*a+100*b +1) / (100*d + 10*c+1)
10 experts max
labels
OK,ProbablyOK,Unknown,ProbablyERROR,ERROR
1)automatic verification -> assigns OK, ? , Not OK
2)Expert verification (up to 10 experts) - assigns (a)OK(expert), (b) Almost OK (expert), (c) Almost Not OK (expert), (d) Not OK (expert)
Each experts assign exactly one label and one of the values a,b,c,d is incremented by one
Q is calculated by the formula above
--
Auto verification
1)inherent coherence of a single origin (e.g. SMILES -structure correspondence)
- select a property that contains SMILES (not the smiles field from chemicals table!)
- compare this property with the smiles field from chemicals table
- compare this property with structure, if chemicals.smiles is null
2)coherence of multiple origins - compare all structures of same idchemical - assign OK if same, notOK if not same
- retrieve a chemical (by idchemical)
- retrieve structures per chemical
- calculate fingerprints per structure
- compare structure fingerprints with fingerprints from fp1024 (per chemical), if different, assign "ERROR", otherwise assign "ProbablyOK"
- (slow) select one structure and compare all other structures (same chemical) with isomrphism test; assign "ERROR" if not same.
3)identificators correspondence between different origins
- the user selects key field (e.g. CAS) and fields to be compared (e.g. structure, names, descriptors)
- select field (e.g. CAS)
- select field to be compared (e.g. structure)
- for all structures having same CAS calculate Q metric
- save metric under e.g. CAS-QA user
+select field (e.g. smiles, structure)
+select field to be compared (e.g. CAS)
+ for all CAS having same structures calculate Q metric - NOK if there are differences
+save metric under e.g. struc-CAS-QA user
+select field (e.g. PubchemCID)
+select field to be compared (e.g. EINECS)
+ for all EINECS having same CID calculate Q metric ok if all einecs are the same, nok if difference
+save metric under e.g. CID-QA user
- auto verification assigns auto labels as above