Error with cleft.threshold in fanc_partner_summary

[yanghelenh]

Running fanc_partner_summary with the cleft.threshold argument set to anything other than 0 returns the following error:

Error:
! Must subset rows with a valid subscript vector.
ℹ Logical subscripts must match the size of the indexed input.
x Input has size 12386 but subscript `res$cleft_score > cleft.threshold` has size 0.
Run `rlang::last_error()` to see where the error occurred.
Warning message:
Unknown or uninitialised column: `cleft_score`. 

Here's the output of rlang::last_trace() after running rlang::last_error()

<error/vctrs_error_subscript_size>
Error:
! Must subset rows with a valid subscript vector.
ℹ Logical subscripts must match the size of the indexed input.
x Input has size 12386 but subscript `res$cleft_score > cleft.threshold` has size 0.
---
Backtrace:
     ▆
  1. └─fancr::fanc_partner_summary(...)
  2.   ├─fancr::with_fanc(...)
  3.   │ └─base::force(expr)
  4.   └─fafbseg::flywire_partner_summary(...)
  5.     └─fafbseg:::flywire_partners_cave(...)
  6.       ├─res[res$cleft_score > cleft.threshold, , drop = FALSE]
  7.       └─tibble:::`[.tbl_df`(...)
  8.         └─tibble:::tbl_subset_row(xo, i = i, i_arg)
  9.           └─tibble:::vectbl_as_row_index(i, x, i_arg)
 10.             └─tibble:::vectbl_as_row_location(i, nr, i_arg, assign)
 11.               ├─tibble:::subclass_row_index_errors(...)
 12.               │ └─base::withCallingHandlers(...)
 13.               └─vctrs::vec_as_location(i, n)
 14.                 └─vctrs `<fn>`()
 15.                   └─vctrs:::stop_indicator_size(...)
 16.                     └─rlang::cnd_signal(new_error_subscript_size(i, n = n, ..., body = cnd_body_vctrs_error_indicator_size))

fanc_partner_summary runs without errors when the cleft.threshold argument is explicitly specified to be 0 in the function call or when it is not specified in the call and defaults to 0. The above error happens whenever cleft.threshold is set to a positive integer; negative values or non-integers fail the corresponding assertion statements and throw different errors.

Also, I've updated my installation for fafbseg and fancr, and the issue persists.

[alexanderbates]

The issue seems to be the that there is no cleft score for FANC synapse data. You can filter synapses by score though. What would a good threshold be?

I have started to implement this in two branches, for fafbseg and for fancr.

If anyone else wants to finish implementing great - I ran out of time today, but will try to over weekend.

[jefferis]

Thanks @yanghelenh. As Alex says there is no cleft.threshold information available for FANC. You could talk to @unidesigner about this. I left the implementation (which is inherited from FAFB) in the hope that it might one day be available.

[jefferis]

@alexanderbates thanks for the work you mention re the plain connection_score. Unfortunately I have no indication that it does anything useful. For example I've looked there is no correlation between connection_score and cleft_score, but I know that high cleft_scores are predictive of true synapses. For example here are the input synapses onto a sample neuron (720575940621039145) plotted.

If I plot synapses above (green) and below (red) a cleft_score of 50 then you can clearly see a preponderance of red synapses in obviously incorrect places (eg the soma and axon shaft).

But I cannot find a connection_score that separates e.g. false synapses on the axon shaft from the arbours. You have generated a validated dataset with Imaan curating FAFB synapses so if you want to see if there is anything about connection_score that is predictive, I would be happy to hear it.

[jefferis]

PS score in flywire_partners() = connection_score as returned by CAVE.

[yanghelenh]

Thanks @jefferis and @alexanderbates for looking into this issue for me. It was clear from the error that cleft.threshold wasn't returning anything, but I didn't realize that it's because this information isn't available for FANC. I'll talk to @unidesigner about whether it could be implemented for FANC, or if there's some other way to filter the predicted synapses by confidence.

[yanghelenh]

Hm... using the flywire_cave_query function of fafbseg to get a FANC neuron's output synapses returns a column called score, but unlike the connection_score, the values seem to range between 13 and 150ish with a mean around 30ish. It seems like the score for FANC predicted synapses might be its own thing.