Abstract
BACKGROUND: A fundamental challenge in quantitation of biomolecules for cancer biomarker discovery is owing to the heterogeneous nature of human biospecimens. Although this issue has been a subject of discussion in cancer genomic studies, it has not yet been rigorously investigated in mass spectrometry based proteomic and metabolomic studies. Purification of mass spectometric data is highly desired prior to subsequent analysis, e.g., quantitative comparison of the abundance of biomolecules in biological samples. METHODS: We investigated topic models to computationally analyze mass spectrometric data considering both integrated peak intensities and scan-level features, i.e., extracted ion chromatograms (EICs). Probabilistic generative models enable flexible representation in data structure and infer sample-specific pure resources. Scan-level modeling helps alleviate information loss during data preprocessing. We evaluated the capability of the proposed models in capturing mixture proportions of contaminants and cancer profiles on LC-MS based serum proteomic and GC-MS based tissue metabolomic datasets acquired from patients with hepatocellular carcinoma (HCC) and liver cirrhosis as well as synthetic data we generated based on the serum proteomic data. RESULTS: The results we obtained by analysis of the synthetic data demonstrated that both intensity-level and scan-level purification models can accurately infer the mixture proportions and the underlying true cancerous sources with small average error ratios (<7 %) between estimation and ground truth. By applying the topic model-based purification to mass spectrometric data, we found more proteins and metabolites with significant changes between HCC cases and cirrhotic controls. Candidate biomarkers selected after purification yielded biologically meaningful pathway analysis results and improved disease discrimination power in terms of the area under ROC curve compared to the results found prior to purification. CONCLUSIONS: We investigated topic model-based inference methods to computationally address the heterogeneity issue in samples analyzed by LC/GC-MS. We observed that incorporation of scan-level features have the potential to lead to more accurate purification results by alleviating the loss in information as a result of integrating peaks. We believe cancer biomarker discovery studies that use mass spectrometric analysis of human biospecimens can greatly benefit from topic model-based purification of the data prior to statistical and pathway analyses.