Analysis of reverse-phase protein microarrays: A comparison of trends in protein expression derived from computerized linear regression, logistic regression, and dose interpolation models

4461

Oncoproteomic characterization of the tumor microenvironment using reverse-phase protein microarrays represents a powerful platform to study alterations in signaling pathways and post-translational modifications both in response to therapeutic regimens and in relation to clinical course. The absence of standardization among proteomic techniques has contributed to difficulty in reproducing proteomic endpoints between groups. There is debate regarding the optimal computational handling of proteomics data. Newly emergent logistic modeling algorithms designed to interpret relative protein expression by image quantitation are being tested. These approaches should be compared to peer-reviewed procedures involving dose interpolation and linear regression. This will advance clinically meaningful oncoproteomic applications. Lysates derived from cell culture and human biopsy specimens were robotically arrayed on nitrocellulose slides in duplicate six-point dilution curves and subsequently stained with validated antibodies. Trends in protein expression values generated from computerized linear regression, logistic regression, and dose interpolation models were compared. Inter-rater reliability, intra-rater reliability, and coefficient of variation among intra-run duplicates were assessed within each computational method. A preliminary examination of trends in protein expression among biologic samples suggests poor correlation between linear regression and logistic regression analyses. Adequate intra- and inter-rater reliability were demonstrated. Comparisons with dose interpolation methods are ongoing. These initial results highlight the importance of properly and carefully applying the appropriate computational methods to proteomic data sets to preserve the integrity of resultant conclusions.