Targeted variant detection using unaligned RNA-Seq reads

(A) Visual description of how km detects the DNMT3A R882 SNP, using k-mers of 31 bp. The input sequence (target sequence given by user) is centered on DNMT3A’s 882nd codon. This sequence is segmented in k-mers to create a linear graph, which represents the search space delimited by the starting and ending k-mers (hatched). (B) A variant will be represented by a new path between the two extremities. This path is found by walking along the linear directed graph and following new (i.e., not seen in the target sequence) overlapping k-mers, queried from a sample’s count table. (C) Schematic representation of an SNP k-mer graph with each path representing the target (lower) and the variant (upper) sequences. (D) Same representation for an ITD variant. (E) Same representation showing the use of a variant target sequence to initiate km’s search. Here, the expected variant is detected when all k-mers that overlap the starting and ending k-mers (in red) are found in the sample’s count table. Also, additional variants could be detected if another path is found (in orange).