Classification of human ovarian tumors using multivariate data analysis of polypeptide expression patterns.

Large amounts of data on quantitative gene expression are generated by procedures such as 2-DE analysis of proteins or cDNA microarrays. Quantitative molecular variation may potentially be used for the development of methods for the classification of tumors. We used here the statistical concepts of principal components analysis (PCA) and partial least square analysis (PLS) in an attempt to type ovarian tumors. Using a set of 170 polypeptides, 22 tumors were used to establish a model ("learning set") for classification into 3 groups (benign/borderline/malignant). Eighteen tumors were then used to test the model. Six of 8 carcinomas and 3 of 4 borderline tumors were correctly classified. Two of 6 benign lesions were correctly classified, 3 were classified as borderline and 1 as carcinoma. We conclude that it may be possible to classify tumors according to their constitutive protein expression profile using multivariate analysis, thus making classification by artificial intelligence a future possibility. Copyright 2000 Wiley-Liss, Inc.