Cell-free DNA samples may have low amounts tumor-derived DNA that may be difficult to detect. ichorCNA has been optimized to detect and quantify the amount of tumor DNA in plasma to answer three primary questions:

1. Is tumor-derived DNA present in the cfDNA sample?

2. How do we decide if the cfDNA sample has sufficient fraction of tumor for follow-up whole exome or deeper genome sequencing?

3. How do we calibrate the depth of follow-up whole exome or genome sequencing?

Additionally, ichorCNA detects large-scale copy number alterations (CNAs), which can be used to characterize the genomic landscape in large cohorts.

4. What are the recurrent CNAs in a cohort?

5. How does the genomic CNA profile of a patient change over time between longitudinal plasma samples?

Tumor fraction (TFx) is defined as the global (genome-wide) proportion of the cell-free DNA sample that is tumor-derived; (1-TFx) is the proportion of non-tumor-derived DNA. TFx is similar to commonly used terms used for bulk tumor analyses - tumor content, tumor purity, etc. The tumor fraction estimates are shown in <sampleID>.params.txt, along with the tumor ploidy, subclonal fraction, and metrics for all solutions considered.
This value is the most important parameter used to address the first 3 questions above.

Benchmarking of ichorCNA using metastatic breast/prostate patient cfDNA and healthy donor cfDNA reveals a lower limit of sensitivity for detecting the presence of tumor to be 0.03 TFx (3%). That is, an ichorCNA-estimate of > 0.03 TFx will reliable indicate the presence of tumor for a cfDNA sample sequenced to ~0.1x whole genome coverage. An estimate of < 0.03 TFx indicates lowly detectable (below 3% but the estimate is less accurate) or absence of tumor-derived DNA. In the benchmark, at a 0.03 TFx cut-off, ichorCNA had a 91% specificity (correctly classified 20/22 healthy donors) and 95% sensitivity (classified 1125/1288 cancer patient mixtures).
When the data quality differs, such as sequencing coverage or overall data variance, and a cancer type that is distinctly different than breast and prostate, manual inspection of cases near the 0.03 TFx cutoff and/or tuning of parameters is recommended.

When selecting cfDNA samples for standard depths of whole exome sequencing (e.g. mean target coverage ~150x), we recommend samples with ichorCNA estimates of > 0.1 TFx (10%). From the benchmarking, ichorCNA correctly estimated samples with > 0.1 TFx for 91% (606/669 patient-healthy donor mixtures) and correctly estimated samples with < 0.1 TFx for 96% (613/641 mixtures).
ichorCNA has been tuned to be conservative and may sometimes underestimate TFx. This was a design decision that this leads to some samples with 0.05-0.09 estimated TFx that may still be suitable for standard depth whole exome sequencing. We recommend users perform manual inspection for a few of these cases in order to consider them for further sequencing.
For samples with lower than 0.1 TFx, additional sequencing can be calibrated to increase the power to detect mutations in whole exomes. images/wes_power_curve.pdf

Samples that have an estimated tumor fraction ~0.03-0.1 may require manual curation to confirm this estimate. Our benchmarking shows that the tumor fraction estimate is most accurate when there is at least one amplification and deletion event spanning more than 100Mb each. This helps anchor the copy number levels correctly at low tumor fractions. Having prior knowledge of the common copy number alterations seen in your specific cancer type is one of the best ways to evaluate the validity of events and subsequently tumor fraction estimates for low purity samples. If a high proportion of CNAs are being called as subclonal, it may be necessary to choose an alternate solution or try rerunning ichorCNA without estimating subclonal status (--estimateScPrevalence FALSE --scStates "c()") since it is difficult to make the clonal/subclonal distinction with low purity samples using coverage alone.

Sometimes ichorCNA will choose a suboptimal solution. Some easy ways to spot a potentially incorrect solution are if:

• A large proportion of CNAs are being called subclonal.
• The majority of data points are brown or red, suggesting a whole genome amplification event.
• There are two distinct copy number levels being called neutral.

These scenarios don't always indicate a suboptimal solution but are commonly seen in them. It is important to check the next best solutions, as ranked by log likelihood, to see if any of them appear to better explain the data. If the solution chosen is suboptimal, the next best solution or two is often the better choice.