Order of input contigs influences the results of combineTCR

[michael-kotliar]

removeNA parameter (I use v1.11.0 installed from GitHub) doesn't work as expected. combineTCR function returns different results depending on the order of the contigs in the input file. We don't see big difference unless we set removeNA=TRUE and/or filterMulti=TRUE.

For example, for input where TRB is before TRA

barcode			is_cell	contig_id			high_confidence	length	chain	v_gene	d_gene	j_gene	c_gene	full_length	productive
AACTGGTCATTAGCCA-1	true	AACTGGTCATTAGCCA-1_contig_1	true		499 	TRB	TRBV9	TRBD1	TRBJ2-1	TRBC2	true		true
AACTGGTCATTAGCCA-1	true	AACTGGTCATTAGCCA-1_contig_2	true		563	TRA	TRAV22		TRAJ17	TRAC	true		true
AACTGGTCATTAGCCA-1	true	AACTGGTCATTAGCCA-1_contig_3	true		507	TRB	TRBV6-4		TRBJ2-7	TRBC2	true		true

the output of combineTCR function will be

barcode			is_cell		contig_id			high_confidence		length	chain	v_gene	d_gene	j_gene	c_gene	full_length	productive
AACTGGTCATTAGCCA-1	true		AACTGGTCATTAGCCA-1_contig_1	true			499	TRB	TRBV9	TRBD1	TRBJ2-1	TRBC2	true		true 

But for input where TRA is before TRB (this also happens when we set filterMulti=TRUE, because in addition to deletion of one TRB we group by barcode and chain)

barcode			is_cell	contig_id			high_confidence	length	chain	v_gene	d_gene	j_gene	c_gene	full_length	productive
AACTGGTCATTAGCCA-1	true	AACTGGTCATTAGCCA-1_contig_2	true		563	TRA	TRAV22		TRAJ17	TRAC	true		true
AACTGGTCATTAGCCA-1	true	AACTGGTCATTAGCCA-1_contig_1	true		499 	TRB	TRBV9	TRBD1	TRBJ2-1	TRBC2	true		true
AACTGGTCATTAGCCA-1	true	AACTGGTCATTAGCCA-1_contig_3	true		507	TRB	TRBV6-4		TRBJ2-7	TRBC2	true		true

The output of combineTCR function will include <NA> for d_gene. Thus, this line will be removed when we set removeNA=TRUE, because instead of checking only TCR1 and TCR2 columns we use na.omit for all columns here.

barcode			is_cell		contig_id			high_confidence	length	chain	v_gene	d_gene	j_gene	c_gene	full_length	productive
AACTGGTCATTAGCCA-1	true		AACTGGTCATTAGCCA-1_contig_2	true		563	TRA	TRAV22	<NA>	TRAJ17	TRAC	true		true

[ncborcherding]

Hey Michael,

Thanks for outlining the problem above - is there a difference in the clone designations i.e. headers CTaa, CTnt, CTstrict, CTgene?

The headers/columns you include above are actually ignored after combineTCR().

Nick

[michael-kotliar]

Hi Nick,

I think all the other columns look ok, but because d_gene has NA, certain cells will be mistakenly removed when setting removeNA=TRUE, and almost all cells will be removed when combining removeNA=TRUE and filterMulti=TRUE. This happens because for filterMulti we group by barcode and chain, making the order of chains to be first TRA then TRB, and as a result d_gene will be NA for all cells except those which have only TRB chain.

[ncborcherding]

Hey Michael,

Apologies for the confusion - I am knee deep working on v2 of scRepertoire (the current main branch).

You may be right that there is unintended removal of contigs using removeNA=TRUE. In scRepertoire v2, combineTCR() output is:

Everything is concatenated and the individual gene calls are not retained, so this should no longer be an issue.

Nick

[michael-kotliar]

Great! Then I'll start using the latest version from the main branch.