Methods and compositions for classification of samples

A method to identify a subject as having a cancer, the method comprising:

• (a) subjecting a first portion of a sample to cytological testing that indicates that said sample is ambiguous or suspicious, wherein said sample is obtained from a subject having or suspected of having said cancer;
• (b) assaying levels of expression of gene expression products in a second portion of said sample from (a), wherein said gene expression products correspond to a set of genes, which set of genes does not include BRAF;
• (c) applying said trained machine learning classifier to said level of expression from (b) to classify said sample as containing or not containing a BRAF mutation at an accuracy of at least about 80%, wherein one or more technical factor variables are removed upon classifying said sample;
• and
• (d) outputting an electronic report that identifies said subject as having said cancer based at least in part on said sample containing or not containing said BRAF mutation.