1. Supplemental Figure 1. Bias-corrected NGS bioinformatics strategies. Paired-end DNA sequencing reveals the sequence of the genomic clone, the sample ID and the unique sequence identifiers in READ_1 and a much shorter READ_2 that determines the probe associated with each clone. The sequence identification tags allow for redundant reads derived from the same original clone to be grouped into sets. By generating consensus unique reads from redundant sets, sequencing errors (red X) are easily identified and removed. True mutation candidates occur when several independent consensus reads all possess the same lesion. The method enriches for sequence information that is 3’ to the capture probe. In cases of opposing probes, regions such as coding exons that are vulnerable to driver mutational events are covered in both orientations. Importantly, the reads from one opposing probe generate sequence coverage of the adjacent probe binding site, providing validation for the probe binding sites used in every experimental sample. For the de novo discovery of gene fusion events, capture probes are tiled across frequently rearranged intronic sequences.

2. AB Supplemental Figure 2: Comparison of targeted sequencing of cfDNA using standard hybridization selection versus Resolution methods. (A) UCSC genome browser depiction of read coverage in the TP53 vicinity. The coverage track representing hybridization selection of cfDNA was created by mapping the reads from sample SRR1197557 of Newman et al 2014. The other coverage track represents unique reads from Sample 511 in this paper. The off-target noise outside of the TP53 gene is higher in SRR1197557 compared with Sample 511. (B) Genome-wide calculation of off-target versus on-target coverage for the two samples. The percentage of read coverage overlapping the declared target regions (widened by an extra 100 nucleotides on each side) was calculated using bedtools.

3. Supplemental Figure 3: Sensitivity of mutant allele detection as a function of dilution. A total of four gene fusions and one indel (orange open squares) and thirteen point mutations (black circles) were assayed at varying mutant allele concentrations (Supplemental Table 2). All 18 lesions were detected in the 2.5%, 1.0%, and 0.4% pools, while fourteen out of the 18 were detected at 0.1%. We expect gene fusions to be more difficult to detect because adequate sequence coverage demands a larger assay footprint than point mutations and probe orientation is unidirectional rather than bidirectional. Consistent with this expectation, the ROS1 fusion in cell line HCC78 and the ALK fusion in cell line H3122 were underrepresented.

4. Supplemental Figure 4: Schematic of predicted genomic junctions for 6 translocations identified in plasma Specimens. 10 RET chr10:43,610,044 1516 KIF5B chr10:32,316,579 KIF5B – RET 1112 13