2.3 Analysis of TCR HTS

Analyses were performed using TCRklass 0.6.0, MiTCR 1.0.3, MiXCR 1.6, IMSEQ 1.0.1, MiGEC 1.2.3 and RTCR 0.3.0, using the default settings. As ’default settings’ for IMSEQ we turned on its clustering based error correction and merging of identical CDR3 sequences with ambiguous segment identification, i.e. ‘-ma -qc -sc’. Before evaluating the performance of each pipeline, non-functional TCR sequences (i.e. those that are out-of-frame or contain a stop-codon) were removed. For the analyses of non-barcoded HTS data (real and simulated), all pipelines, with the exception of MiXCR which uses its own reference, were run with the germline reference sequences of MiTCR. For the analysis of barcoded HTS data, RTCR was run with the V(D)J reference sequences of MiGEC. To compare the error correction of MiGEC and RTCR, the latter was run on the sequences resulting from the Checkout utility of MiGEC so that both had the same starting point. MiGEC was run with an ‘overseq’ threshold of 5 (i.e. discarding UMI groups with fewer than five reads).

Both TCRklass and IMSEQ pipelines report identical CDR3 sequences with different VJ combinations by default which can inflate the false positive rate. To make the reporting equivalent among the pipelines, we collapsed these sequences and summed their counts. Collapsing these sequences had only a minor positive effect on the precision of TCRklass and IMSEQ and no effect on the recall.