You are here: Technology >> Positioning

Positioning

Where Equinox Pharma has positioned itself

INDDEx™

Technology and methodology behind INDDEx™

Advantages

Advantages of using INDDEx™ over other technologies

Targets

Results of INDDEx™ application to various targets

Publications

Publications relating to Equinox Pharma technologies

INDDEx™ technology

Equinox’s primary technology is INDDEx™ (Investigational Novel Drug Discovery by Example). INDDEx™ is primarily a ligand-based approach to virtual screening, but it can also include receptor-based information. INDDEx™ is based on a novel logic-based machine learning approach generating rules from quantitative structure activity relationships. INDDEx™ uses the patented SV-ILP (Support Vector Inductive Logic Programming) methodology, in which a rule learning engine generates logic-based rules which are quantified with support vector machine methods. The schematic below illustrates how these technologies are integrated to produce a predictive model of activity from a training dataset.

 

A graphical representation of the INDDEx™ process

Left-click on parts of the diagram for an explanation of the technology.
Input molecules Fragmentation methods Machine learning method Rule-based model Screening Database Output molecules

Input molecules

The input molecules form the training dataset for the INDDEx™ program. INDDEx™ can learn from a small or large number of active and inactive molecules, and can be used with either binary classification or with regression from the individual activity levels of the molecules for greater accuracy.

Fragmentation methods

The molecules are dissected into chemically-relevant fragments using our proprietary fragmentation methods. The molecular structures of the active and inactive examples from the training dataset are fragmented into logical statements for input into the machine-learning program.

Machine learning method

INDDEx™ technology uses machine-learning to generate a set of logic-based rules about the relative positions of structural fragments in the ligands that are responsible for ligand activity. These rules can be easily used by medicinal chemists to understand the mechanism of activity, and help guide the hit-to-lead process.

Rule-based model

A rule-based filtering model is created from the logical rules generated in the previous step. Each rule gives the relative positions of two substructural fragments. Regression is used to combine the rules to yield a filtering model. Methods include both Support-Vector Machines and Partial Least Squares. The integration of logic-based rules with support-vector regression is our patented SV-ILP technology (Patent number 8126823).

Database

The most commonly used database our customers employ is the ZINC database, a database of commercially-available molecules that we use for virtual screening. Our customers can also supply their own proprietary or other databases for virtual screening.

Screening

The filtering model is used to scan the database. INDDEx™ scans either all molecules in the database, or the subset of all purchasable molecules. We can scan any molecular database on request.

Output molecules

The output of INDDEx™ is a ranked list of molecules predicted to have high activity. A particularly powerful feature of INDDEx™ is scaffold-hopping, where the model often identifies molecules that are chemically distinct from any in the training dataset. Equinox has shown that the output molecules are more chemically diverse than those achieved by other commercially-available processes. (See INDDEx™ advantages).