Bayesian hierarchical inversion for mass spectrometry. Application to discovery and validation of new protein biomarkers

BHI-PRO est accrédité par Lyonbiopôle

Identity card

Global budget: 1 824 k€

Public funding: 838 k€

Public funders: ANR, Agence Nationale de la Recherche

Call for projects: ANR (ANR2010 - BLANC)

Year start: 2010

Completed project (2014-06)

Accredited by the French competitiveness cluster Lyonbiopôle

Strategic business area: Human Medicines, In vitro Diagnostic, Medical Technologies

Application fields: Health & Environment, Oncology

Technological approaches / Keywords: Bioinformatics / Software, Omics Technologies, Proteomics - Mass spectrometry - Signal processing - Biostatistics

Stage of development at the beginning of the project: Basic research


To reduce the impact of technological variability on protein quantification using MS-based clinical proteomics, we propose to introduce a relevant hierarchical modeling of MALDI and MRM chains within a Bayesian statistical framework. The new Bayesian Hierarchical Inversion algorithms are relying on two advances: the association between proteomic analysis, Bayesian inverse problems, stochastic sampling from one side and the convergence between biostatistics and inverse problem from another side. We evaluated the statistical performances on protein quantification, sample classification and biomarker selection. Evaluation has been carried on synthetic and patient samples within oncology studies. 


To automatically operate a mass spectrometry data analysis To shorten biomarker validation studies with respect to immunoassays (ELISA) To increase robustness of discovery and validation studies on a MALDI (CLIPP) and a MRM platform (bioMérieux) by controlling technological variability To design a Bayesian framework combining hierarchical mixture models, probability distributions, and stochastic sampling inversion algorithms To evaluate the statistical performances for protein quantification, sample classification and biomarker selection To study synthetic and patient sample cohorts in oncology To combine in a single research project Bayesian inversion, biostatistics, and proteomics platforms 

Innovative assets
Actual results

Innovative assets

For MALDI mode, the study has demonstrated that BHI-PRO pre-processing was able to reduce the impact of technological variability on reconstructed profiles when keeping biological variability. Its quantification performances are better than the usual algorithm on the platform. In MRM mode, we have demonstrated that the automatic BHI algorithm was able to control automatically the influence of technological variability when the quality of calibration was satisfactory. Its performances for classification are the same or even better than the operator-supervised Non-Linear Processing method currently used. This opens the way towards processing larger cohorts and developing automated diagnostic tests.

En poursuivant votre navigation sur notre site, vous acceptez l'utilisation des cookies et la collecte de vos données et informations personnelles par Lyonbiopôle, dans les finalités de mesurer le trafic sur le site Web, de fournir des statistiques et de vous proposer des contenus adaptés à vos centres d’intérets. Pour exercer vos droits d'accès, de rectification, d'opposition, de suppression et de portabilité, conformément au règlement général sur la protection des données (UE n°2016/679), vous etes informés que vous pouvez envoyer votre demande à Plus de détails sont disponibles en cliquant ici J'accepte