Additional file 5: of Machine learning enables detection of early-stage colorectal cancer by whole-genome sequencing of plasma cell-free DNA WanNathan WeinbergDavid LiuTzu-Yu NiehausKatherine AriaziEric DelubacDaniel KannanAjay WhiteBrandon BaileyMitch BertinMarvin BoleyNathan BowenDerek CreggJames DrakeAdam EnnisRiley FransenSigne GafniErik HansenLoren LiuYaping OtteGabriel PecsonJennifer RiceBrandon SandersonGabriel SharmaAarushi St. JohnJohn TangCatherina TzouAbraham YoungLeilani PutchaGirish HaqueImran 2019 Figure S4. Non-linear relationship between the total number of samples used for training and sensitivity at 85% specificity for colorectal cancer detection. The method was trained again with k-fold, except the number of training samples per fold was downsampled. The lower numbers are comparable to those available for balanced k-batch and were used to investigate decreased classifier performance due to smaller sample sizes in training. (DOCX 60 kb)