A1101
Title: IsoBayes: A Bayesian approach for single-isoform proteomics inference
Authors: Simone Tiberi - University of Bologna (Italy) [presenting]
Jordy Bollon - Italian Institute of Technology (Italy)
Abstract: Inferring proteins is a crucial step in biomedical research. At present, proteins are indirectly measured via peptides. However, this process is noisy because most peptides are shared across multiple proteins; furthermore, peptides may also be erroneously detected. As a consequence, studying proteins is challenging, and inferences can be inaccurate. IsoBayes, a novel Bayesian statistical method, is described for protein inference. The goal is to disentangle the biological variability (of interest) from the technical noise arising from the measurement process (nuisance). To this aim, a two-layer latent variable approach is designed where: first, it is sampled if a peptide has been correctly detected, and second, the abundance of such selected peptides is allocated across the protein(s) they are compatible with. This framework enables us, starting from peptide-level data, to recover protein-level information (i.e., presence and abundance). Furthermore, proteomics and transcriptomics data are integrated to enhance the information available. In order to validate the approach, comprehensive benchmarks are designed based on simulated and real datasets, where IsoBayes displays good sensitivity and specificity when detecting proteins and where its estimated abundances highly correlate with the ground truth. Importantly, the method is flexible and works with peptide identifications obtained by any proteomics tool, and it is distributed open-access as a Bioconductor R package.
