It's unclear to me what the pattern is for what columns are being selected here. Can you tell me what the particular pattern or strategy for these columns being selected is?

Maybe we can come up with a more succinct way of selecting these if there's a pattern.

For this, since we are specifically looking for in-frame (Fusion_Type), kinase (Gene1A_anno, Gene1B_anno), whether a reciprocal exists (reciprocal_exists ), and fusions with kinase domains retained (DomainRetainedGene1A, DomainRetainedGene1B), these columns required for LGAT fusion compilation. I added the left and right breakpoint columns to make sure that I was seeing unique fusions, although the rows may be unique enough with the combination of columns above. Let me check on this.

Looks like the number of unique fusions without breakpoints == 265 and with breakpoints == 300, but the results are identical. I will remove those two columns!

I'm thinking this chunk can still be simplified. In particular, I'm noticing that five_prime_domain_lost is from allFuseLGAT and five_prime_reciprocals is from five_prime_domain_lost but then we are adding back info from allFuseLGAT.

So basically both five_prime_domain_lost and five_prime_reciprocals are based on allFuseLGAT but we have extra manipulation steps. I think we can do this more directly.

As I'm trying to make this shorter, I'm a bit confused on these steps.

I tried to boil it down to what looks like the essential steps to get five_prime_kinase_keep, but as of now it does not get the same result. I think part of this is I still don't really understand all that is happening in this chunk. Can you take a look at my attempt as well evaluating your original code to see how we can make these steps more efficient and clear?

In particular I know the differences have to do with the paired up "select() then distinct()" steps that are done a few timees, but its unclear to me why each of these "select() then distinct()" steps are necessary.

five_prime_kinase_keep <- allFuseLGAT %>%
  # Make reciprocal variable
  mutate(reciprocal = paste(Gene1B, Gene1A, sep ="--")) %>%
  filter(# Keep the in-frame reciprocals which have the kinase domain in tact.
        grepl("Kinase", Gene1A_anno) & reciprocal_exists == "TRUE" & DomainRetainedGene1A == "No" |
        # Also keep select the in-frame reciprocals which have the kinase domain in tact.
        Fusion_Type == "in-frame" & DomainRetainedGene1B == "Yes") %>% 
  # Select to which variables we want here
  select(Sample, FusionName = reciprocal) %>% 
  distinct()

Ahh

        grepl("Kinase", Gene1A_anno) & reciprocal_exists == "TRUE" & DomainRetainedGene1A == "No" |
        # Also keep select the in-frame reciprocals which have the kinase domain in tact.
        Fusion_Type == "in-frame" & DomainRetainedGene1B == "Yes")

This filtering you have here is filtering on the frame-ness/domain retention of the 5' kinase fusions and not their reciprocals (3' kinase fusions), which is why I was making that variable, then re-joining five_prime_reciprocals with the allFuseLGAT.

What I am trying to do is find any 5' kinases without the domain retained, but then go back to that original dataframe allFuseLGAT and collect the annotations of their reciprocal fusions (ie 3' kinase fusions' frameness and domain retention), assess whether those are in-frame and retain the kinase domain, then if there is a reciprocal that satisfies that, join both the 5' kinase fusion dataframe (domain lost) and 3' kinase fusion dataframe (inframe/domain retained).

Actually, I can remove:

# Add back the reciprocal of the reciprocal (the 5' kinase fusion) and keep this list
five_prime_kinase_keep <- five_prime_reciprocals %>%
  mutate(reciprocal = paste(Gene1B, Gene1A, sep ="--")) %>%
  select(Sample, FusionName = reciprocal) %>%
  distinct()

and use:

# Then, select the in-frame reciprocals which have the kinase domain in tact. Retain 5' kinase fusion information and update 3' fusion column name to FusionName for merging with allFuseLGAT.
five_prime_kinase_keep <- five_prime_domain_lost %>% 
  select(Sample, five_prime_kinase = FusionName, FusionName = three_prime_kinase) %>%
  left_join(allFuseLGAT, by = c("Sample", "FusionName")) %>%
  filter(Fusion_Type == "in-frame" & DomainRetainedGene1B == "Yes") %>%
  select(Sample, FusionName = five_prime_kinase) %>%
  distinct()

distinct() is needed because in some cases, there are multiple breakpoints for each fusion and sometimes multiple can have the same frame-ness, so this reduces them to one fusion call per sample.

updated with commit - @cansavvy let me know if this clarifies!

Okay! Thanks for those updates! I think this looks good!

Okay! Thanks for those updates! I think this looks good!