Xavier Robin - Research

From images acquired with an Opera high-content screening (HTC) microscope, I extract morphological features of the cells present in the images, such as shape, textures and intensities. The resulting Quantitative Phenotypic Signatures (QTP) correspond to a highly dimensional feature space that has to be reduced with dimensionality reduction techniques before machine learning can be applied. An additional challenge is the very high sample size - HTC can image millions of cells, and most dimensionality reduction algorithms currently available are not able to deal with such a high number of data points.

Similar to weather forecasts, biological forecasts can be developed to predict cell behaviour following a perturbation (or cue). Super-computing facilities are required to model the complex non-linear networks of interactions and their effects. Such models integrate genetic (sequencing), expression (proteomics) and phenotypic (imaging) data with neural network algorithms, and allow to derive information on the signaling network architecture.

Massive amounts of mass spectrometry data are generated in the lab with state-of-the-art instrumentation (Thermo Scientific Orbitrap Q-Exactive and Fusion) and high-throughput workflows. I am developing computational tools to to deal with this data, and better statistical models able to take into account the experimental uncertainties of the datasets in a statistically correct manner. In the long term, this will allow us to move away from FDR (false discovery rates) and arbitrary cut-offs. Ensuring the correctness of the uncertainties also appears to be of critical importance for the global models developed in the group.

In a longer-term perspective, I am interested in developing personalized network-based treatments for cancer. With biological forecasting models, it becomes possible to predict which drug (or several drugs) must be given to the patient (and in the case of multiple drugs, in which order (time-staggering)), in order to either restore a proper function of the cell signaling network or kill cancer cells. These treatments result from a deep understanding of how the cell signaling network is wired, and how it will react to a given treatment.

tment.

Publication

Robin Creixell, Radetskaya et al. Personalized network-based treatments in oncology, Clinical Pharmacology and Therapeutics. 2013;94(6):646-50.

I am also involved in various bioinformatics projects of the lab, especially those involving the development of web interfaces, statistical analysis, quantitative analysis and parallel computing.

Publication

Horn, Schoof, Kim et al., KinomeXplorer: A Novel Integrated Framework for Kinome Biology Studies, Nature Methods. 2014;11(6):603-4.