BACKGROUND: Matrix-assisted laser desorption ionization time-of-flight (MALDI-TOF) proteomic signature might be of interest for the early detection and staging of hepatocellular carcinoma (HCC). However, published procedures have been criticized for the lack of data about analytical reproducibility, and the use of inadequate data processing. METHODS: MALDI-TOF profiling of peptides bound to serum albumin ("albuminome") was performed using 90 μL of serum from 45 study subjects (HCV-related cirrhosis, small, unifocal HCCs and advanced HCCs). To overcome the large intra-sample variability, a Quality Assurance protocol was implemented, and 4-8 samples for each subject were processed and analyzed. Overall, 522 subject samples and 299 quality-control spectra were analyzed. A machine-learning approach (Random Forest) was applied to analyze the data sets. RESULTS: Mean intra-sample coefficient of variation (CV) of the analytical procedure was 17.6%-30.0%; inter-subject CV was in the range 48.8%-71.3% among the three study groups. The Random Forest procedure correctly classified 433/522 "patient samples" and 295/299 "reference samples"; 43/45 patients were correctly classified following this approach. CONCLUSIONS: Our data suggest that, notwithstanding the large analytical variability found, multiple proteomic profiles obtained from each subject can differentiate cirrhosis with and without HCC, and HCCs with and without vascular invasion, warranting further investigation in a prospective setting.
BACKGROUND: Matrix-assisted laser desorption ionization time-of-flight (MALDI-TOF) proteomic signature might be of interest for the early detection and staging of hepatocellular carcinoma (HCC). However, published procedures have been criticized for the lack of data about analytical reproducibility, and the use of inadequate data processing. METHODS: MALDI-TOF profiling of peptides bound to serum albumin ("albuminome") was performed using 90 μL of serum from 45 study subjects (HCV-related cirrhosis, small, unifocal HCCs and advanced HCCs). To overcome the large intra-sample variability, a Quality Assurance protocol was implemented, and 4-8 samples for each subject were processed and analyzed. Overall, 522 subject samples and 299 quality-control spectra were analyzed. A machine-learning approach (Random Forest) was applied to analyze the data sets. RESULTS: Mean intra-sample coefficient of variation (CV) of the analytical procedure was 17.6%-30.0%; inter-subject CV was in the range 48.8%-71.3% among the three study groups. The Random Forest procedure correctly classified 433/522 "patient samples" and 295/299 "reference samples"; 43/45 patients were correctly classified following this approach. CONCLUSIONS: Our data suggest that, notwithstanding the large analytical variability found, multiple proteomic profiles obtained from each subject can differentiate cirrhosis with and without HCC, and HCCs with and without vascular invasion, warranting further investigation in a prospective setting.
Authors: Ronald J Holewinski; Zhicheng Jin; Matthew J Powell; Matthew D Maust; Jennifer E Van Eyk Journal: Proteomics Date: 2013-01-29 Impact factor: 3.984
Authors: Paul Dowling; Valentina Palmerini; Michael Henry; Paula Meleady; Vincent Lynch; Jo Ballot; Giuseppe Gullo; John Crown; Michael Moriarty; Martin Clynes Journal: BBA Clin Date: 2014-09-16