Team's publications / Organization Chart

Acquiring knowledge of the complete pharmacology profile has inspired new strategies to predict and to characterize drug-target interactions in order to improve the success rates of current drug discovery paradigms, i.e. increase the efficacy and reduce toxicity and adverse effects. With the production of massive amount of omics data (proteomics, transcriptomics, genomics, …) and the increase accessibility for academic to large biological and clinical data, the development of computational approaches to evaluate the drug safety and drug pharmacology is now possible.
Our team focuses on the analysis of the interactions between small molecules and proteins as well as the assessment of their effect on systems biology, i.e. Systems Chemical Biology. Our main objective is to gain a better understanding of the mechanism of action of small molecules across multiple scales of complexity in human health, with the development of computational approaches applied to pharmacological profiling.

Three complementaries tasks are driven in our team :
i) Chemical space (chemoinformatics, drug-design)
ii) Target space and pockets (structural bioinformatics)
iii) Biological space (data integration, systems pharmacology, pharmacogenomics)

Some recent studies outlined the need to combine protein target information (3D) with their binding sites (protein pockets or channels) together with chemical information (small molecules), in order to characterize and to predict the target-ligands interactions and the profiles of the partners of these interactions. Furthermore, the information obtained from i) and ii) are then combined with the biological and clinical data collected in iii)

(i) Chemical space (chemoinformatics, drug-design)

Ligands space profiling: optimal encoding of bioactive compounds molecular structures into relevant descriptors in order to analyse molecular diversity. Several softwares compute thousands molecular descriptors or fingerprints, but their physical meaning is often unclear. Some crucial ones are lacking: it is a bottleneck to get robust QSAR results and get reliable pockets-ligands interactions models.
We regularly develop new descriptors for the community, see http://petitjeanmichel.free.fr/itoweb.petitjean.freeware.html and references cited. A recent study concluded that spherical ligand shapes were physically unrealistic and that cylindrical shapes were better: either minimal height cylinders (relevant for most molecules), or minimal radius cylinders (Petitjean M., Appl. Alg. Eng. Comm. Comp., 2012), relevant for rather elongated molecules. This approach leads to select possible trajectories of potential active substrates in protein channels (Benkaidali et al., Bioinformatics 2014), which are useful in a virtual screening context.

 

(ii) Target space and pockets (Structural bioinformatics)

It was pointed out that the key properties of a good drug target had not been discussed sufficiently or well-defined and that «druggability» (i.e. it can be bound and modulated by a drug-like) have to be taken into account early in the drug discovery process. Indeed despite advances in both experimental and computational fields, around 60% of drug discovery projects fail because the target is not druggable.
Our main focus is to optimally characterize 3D targets and their pockets and to identify therapeutically-relevant drug targets that meet the criteria of being druggable. To simplify 3D structure analysis, we have developed a Structural Alphabet (SA-tools, Camproux et al, J. Mol. Biol. 1999, 2004) encoding any 3D structure into a 1D sequence of structural letters, relevant to analyze and to compare complete 3D structures or local fragments such as binding sites (Regad et al, BMC Bioinformatics, 2011, NAR, 2011).
This can be used to analyse functional site as it is expected their comparison can detect off-targets. In the aims to enhance understanding of the pocket, we develop tools to (i) estimate pockets (Benkaidali et al., Bioinformatics 2014), (ii) characterize pockets by original descriptors (Pérot et al, DDT 2011), (iii) cluster and compare pockets. Resulting descriptors are selected and statistically combined to predict «druggability». Our results to distinguish druggable pockets are very efficient using a set of «free» descriptors and different pocket estimation.

 

Pocket-ligand joint space analysis and statistical profiling (proteochemometrics, chemogenomics)

If a pocket is druggable, we can try to predict which geometrical and physico-chemical windows are required by the ligands (ligand profiles) to correspond to the properties of the pockets. The descriptors extracted from the target space can be used to propose target similarity and matched with ligand similarity extracted from the ligand space in order to predict potential interactions. By the statistical modeling of pocket-ligand pairs and the use of correlated pairs descriptors, a correlation between the nature of the pocket and its propensity to bind some ligand profiles and conversely, a correlation between the nature of the ligands and its propensity to bind some pocket profile has been established using machine learning methods (Pérot et al. Plos One 2013). This pocket-ligand profiling modeling aims to profile prediction from one partner of the interaction could open the ways to the prediction of interactions with off-target, toxic effects, side effects or facilitating the design of a small molecule for a given pocket.

 

(iii) Biological space (data integration, systems pharmacology, pharmacogenomics)

 

 

 

Overall, our objective is to combine all theses data collected through several repositories and predicted by our computational approaches in order to provide insights into the drug pharmacology, drug safety and differentiation of patient drug response toward a personalized medecine.