Iterpretation of and candidates selection according to rank and/or confidence

[vedifredi]

I'm a little bit confused regarding selection of the best candidate from multiple ranked ligand poses. I see for instance these result files after a DiffDock run with default settings and have a couple of questions:

rank1.sdf rank1_confidence0.34.sdf rank2_confidence0.24.sdf rank3_confidence0.22.sdf rank4_confidence0.20.sdf rank5_confidence0.17.sdf rank6_confidence0.15.sdf rank7_confidence0.07.sdf rank8_confidence-0.04.sdf rank9_confidence-0.04.sdf rank10_confidence-0.33.sdf

• Obviously, the confidence also take negative values (rank8-10), is that true?
• Are the smaller or larger confidence associated with higher result confidence?
• Does the rank (no. 1-10) have higher priority than the confidence value or vice versa regarding the selection of favorable candidates?

Strangely, I sometimes observe confidence values that monotonically increase from rank1 to rank 10 (with rank1 confidence close to 0, above example) and sometimes the opposite, monotonically decreasing (with rank1 confidence close to 1). But never (so far) a unsorted (non-monotonic) relation between confidence and rank. Isn't that possible or kind of a bug?

[vedifredi]

OK, I was confused by the hyphen/minus symbol but see things clearer now:

• Confidence values can be < and > 0
• The ranking is always sorted according to confidence values
• Large confidence value -> more trust (better candidate)

[srilekha1993]

Is it fine to have multiple confidence for a particular rank like below rank2_confidence-0.59.sdf
rank2_confidence-0.65.sdf
rank2_confidence-0.68.sdf