Leveraging MD 2.0 to support proteomics-based biomarker discovery

Typical proteomics-based Biomarker discovery approaches aim to define a small subset of biological markers able to predict biological groups of interest. While biomarker discovery spans a broad range of applications, a key use is the detection of markers able to signal the early onset of a disease, increasing the potential to inform and highly improve the treatment of patients.

Despite the potential of detecting early diagnosis markers, the number of well-validated and clinically established markers remains low. A biomarker discovery workflow requires careful consideration when defining the experiment design, the data preparation, the downstream bioinformatics analysis, and integration with knowledge databases for external validation. Furthermore, validation of the defined set of markers across independent datasets is required to prove their quality and biological reproducibility.

Therefore, once proteins or peptides have been identified and quantified, a number of quality control, statistical analyses, and validation steps routinely follow. However, these are often developed for and by computational biologists. Mass Dynamics 2.0 (MD 2.0) exists to make current and new computational methodologies accessible to all researchers in an intuitive modular, web, and cloud-based collaborative environment. In addition, MD 2.0 enables seamless results comparison and multi-dataset management so that a researcher can easily organize and access their multiple discovery and validation datasets and analyses in one place.

Leveraging publicly available cancer proteomics datasets, we showcase the end-to-end reproducible biomarkers discovery analysis suite in MD 2.0. This includes: 1) interactive modules for quality control, including PCA, t-SNE, and heatmaps to verify the similarities between the biological groups of interest. 2) Different statistical approaches to define a subset of discriminative markers between conditions of interest using independent approaches. These methods include differential expression analysis, discriminant analysis, and penalized logistic regression. 3) Interactive upsets plot to easily define intersections between sets of interest and ROC curves to validate the predictive power of one or subsets of the defined markers.

The workflow presented here highlights the ability to leverage MD 2.0 as a platform for guided assessment of multi-proteomics data to improve the confidence in identifying biomarkers for further assessment. The platform provides a collaborative space that improves the ability for research teams to make key insights into complex proteomics data and lowers the ability for all stakeholders to access and compare various processing approaches and the influence of metadata on results.