Data Integration in Drug-Target Interaction Prediction for Drug Discovery and Drug Repositioning

Approved drugs have favorite or validated pharmacokinetic properties and toxicological profiles, and the repositioning of existing drugs for new indications can potentially avoid expensive costs associated with early-stage testing of the hit compounds. In recent years technological advances in virtual screening methodologies have allowed medicinal chemists to rapidly screen drug library for therapeutic activity against new biomolecular targets. Although the combination of virtual screening and repurposing potentially represents a very efficient and cost-effective strategy in drug discovery, however no perfect computational software for prediction is available. In fact, each approach considers just specific typology of data, disregarding other important information. The goal of my project is to reduce the rate of false negative and false positive, and improve the prediction power in drug repositioning, by fusing different approaches. During the next four years I am going to study and apply structure-based, ligand-based and network-based techniques for predicting the interaction between known compounds and different targets, and I will unify all the results with data-integration methods, in order to obtain more accurate results. The protocol will be applied on several real cases, in order to find new hits for interesting targets, such as some kinases, involved in the metastasis and B-cell activation.

Data integration is a mathematical tool able in-depth insights into processes and can be performed as meta-analysis to integrate heterogeneous data derived from different experiments. In this specific case, matrix factorization will be used for combining different kind of results, derived from three different predictive approaches:

LIGAND-BASED: ligand similarity is assessed by comparing 3D and 2D fingerprinting of the structures of unrelated known compounds, in order to find common characteristics which could lead the binding;

STRUCTURE-BASED: consists in the comparison of pharmacophore based fingerprinting of the binding sites, since the assumption that similar binding sites can binds similar ligands;

NETWORK-BASED: prediction based on the study of network of interactions among proteins, compounds and genes, built on known information, predictions and differential expression of functionally related genes.