How to deal with binning artifacts in the database?

[alephreish]

What are the recommended ways of dealing with contamination in the database? A scaffold closely related to the query automatically yields a high-scoring match, yet if its taxonomic identification is based on a binning artifact, this results in incorrect identification - even if there are matches to other, correctly identified but more remotely related scaffolds in the database.

Take e.g. this contig. It's an archaeal contig that ended up in a flavobacterial bin. The contig is present in gtdb226db and thus receives a high-scoring identification when searched against it:

#query_id  name               taxID      query_length  score    rank     lineage                                                                                                                             taxID:match_count
1          JARRKS010000295.1  258006470  4863          0.86387  no rank  d_Bacteria;p_Bacteroidota;c_Bacteroidia;o_Flavobacteriales;f_Flavobacteriaceae;g_DATGBV01;s_DATGBV01 sp037387865;-_GCA_037387865.1  258006470:464 

Is there a way to exclude certain genomes (e.g. based on its taxID) or scaffolds from the search? One strategy I envision is to do two rounds of classify searches: once against the complete database and the second time excluding the genomes matched in the first round. This would help to deal with singleton binning artifacts as in the case of JARRKS010000295.1.

[jaebeom-kim]

Thank you for providing this interesting point.
I think excluding reference sequences using tax ID can be implemented.
However, I'm not sure it can solve the binning artifacts problem.
How easily user can know that certain genomes are contaminated or not?
In other words, you can exclude somethings you know, but there may be other contamination you don't know.

• Could you explain further you idea? It sounds interesting, but I'm not sure I understood it correctly. Do you want to know the second best hit at the end?

[alephreish]

I'm not sure at all how to solve the issue of MAG contamination for a general case. I can imagine recruiting CheckM contamination scores to weigh the genomes and the lengths of the matching scaffolds to weigh individual scaffolds.

It would be already useful to allow the user to blacklist particular genomes and/or scaffolds. The user can then decide based on their criteria what to exclude (based on CheckM scores, scaffold length or some other considerations).