Parameter tuning and settings

Low tumor content samples (early stage disease)

For samples that are expected to have lower than 5% tumor fraction, it may be helpful to modify the default settings to improve parameter estimation. It is recommended to sequence to higher coverages (> 1-5x) for these types of samples. For samples with less than ~0.5% expected tumor fraction, we recommend standard depths of whole genome sequencing (e.g. > 20x).

1. Initialize tumor fraction parameter
   --normal "c(0.95, 0.99, 0.995, 0.999)"
   Initialize the non-tumor (1 minus tumor fraction) to expected values, such as 5%, 1%, 0.5%, 0.1%. ichorCNA will still estimate the tumor fraction but having these initial starting values can help the EM step find better global optima.

2. Set initial ploidy to diploid
   --ploidy "c(2)"
   It will be difficult to predict the ploidy value for low tumor fraction cases.

3. Reduce number of copy number states
   --maxCN 3
   Reducing the state space will help reduce complexity. If you know from a prior sample (e.g. tumor biopsy) that there are large high level copy number events, you can set this to 4.

4. Do not account for subclonal copy number events
   --estimateScPrevalence FALSE --scStates "c()"
   Subclonal events are difficult to detect for low tumor fraction, these can we turned off.

5. Train and analyze autosomes only
   --chrs "c(1:22)" --chrTrain "c(1:22)"
   Exclude chrX in the analysis and training to reduce complexity.