After DNA isolation on a Maxwell® 16 Research (Promega Corporation, Madison, USA) as recommended in the manufacturer’s protocol, all tumor samples were sequenced using a small panel of 22 genes and 92 amplicons covering hotspots characteristic for NSCLC (Colon Lung v2 AmpliSeq Panel by Thermo Scientific, Waltham, MA, USA).
Non-synonymous mutations with a coverage above 500 reads and an allelic frequency above 3.0% were included into the analysis. Variants with an allelic frequency below 3.0% were filtered out and regarded as artifacts due to formalin fixation. Considering the percentage of tumor cells, the mutations validated needed to be detectable in at least 10% of the tumor sample.
The influence of mutations on proteasomal cleavage was predicted by the machine learning tool NetChop 3.1 [43 (link), 44 (link)]. The binding of the resulting epitopes to MHC class I was subsequently simulated by NetMHC 4.0 [45 (link), 46 ], also based on convolutional neural networks. The whole procedure is described in detail in our previous works [34 (link)].
Non-synonymous mutations with a coverage above 500 reads and an allelic frequency above 3.0% were included into the analysis. Variants with an allelic frequency below 3.0% were filtered out and regarded as artifacts due to formalin fixation. Considering the percentage of tumor cells, the mutations validated needed to be detectable in at least 10% of the tumor sample.
The influence of mutations on proteasomal cleavage was predicted by the machine learning tool NetChop 3.1 [43 (link), 44 (link)]. The binding of the resulting epitopes to MHC class I was subsequently simulated by NetMHC 4.0 [45 (link), 46 ], also based on convolutional neural networks. The whole procedure is described in detail in our previous works [34 (link)].
Free full text: Click here
