I am using the copy number data from the TCGA to observe focal and broad copy number events. After downloading the CNA data I noticed the segmentation allows for segments with as little as two probes. This appears to lead to very noisy data that doesn’t agree with focal CNA studies in literature, and likely due to setting the segmenter to allow 2 or more probes per segment. I cannot find a reference for what is the usual minimum probes used to easily filter these spikes out (I know GISTIC2.0 uses 4 as a default for join segment, but there is no justification beyond leaving it to the user). Any advice or input is welcomed and appreciated!