Question: technical definition of perfect, strict, and loose

[alimayy]

Hi there,

Thanks for developing RGI.

Regarding the perfect, strict, and loose hits, Jia et al. 2017 mentions:

The RGI currently supports two detection model types (Protein Homolog and Protein Variant) and analyzes sequences under three paradigms—Perfect, Strict, and Loose (a.k.a. Discovery). The Perfect algorithm is most often applied to clinical surveillance as it detects perfect matches to the curated reference sequences and mutations in CARD.

Could you provide a more technical definition of these terms? For instance for 'perfect', I understand from https://github.com/arpcard/rgi/issues/101 that the self-mapping bit-scores determine the 'perfect' hits. Could you provide more info on how 'strict' and 'loose' thresholds are chosen?

Thanks in advance, Ali

[raphenya]

@alimayy As an example, Blasting AXC-1 on the CARD database produces the results shown. The bit-score chosen will be 500, as it will capture genes similar to AXC-1.

• A Perfect hit to AXC-1 has all amino acids matching CARD reference.
• A Strict hit to AXC-1 will be any gene with at least bitscore of 500.
• A Loose hit to AXC-1 will be any gene with bitscore below 500.
• All genes in CARD go through the same processes.

[parwa28]

Can you please tell me how exactly I can use one of the detection models?

[agmcarthur]

@alimayy these detection models are used by CARD's Resistance Gene Identifier software: https://github.com/arpcard/rgi (contains full documentation). There is an online portal at https://card.mcmaster.ca/analyze/rgi. Also described in CARD's latest publication: https://pubmed.ncbi.nlm.nih.gov/36263822/