JPS: The Science Behind Acelot’s Leads

Acelot applies patented graph-mining algorithms to the chemical space in order to identify structural properties relevant to a desired activity or binding. At the core of these algorithms is the so-called “Joint Pharmacophore Space” (JPS), a collection of all 3D pharmacophoric point constellations from multiple ligands and/or multiple targets.

Each molecule is analyzed and its pharmacophoric constellations are determined. These constellations are then clustered together with other molecules’ constellations based on their 3D similarity:

workflow

The molecules added to the JPS can be for the same target or for different targets. In the example above, molecules for target A and B were added to JPS. Some clusters contain constellations from both classes of molecules and some clusters contain predominantly constellations from ligands for target A (orange) or target B (blue). This statistical over- or under-representation of constellations in a cluster means that the corresponding constellation is likely contributing to the binding in some way.

This property can now be used to predict whether other molecules may bind to the same target, to a specific set of targets, or to some targets and not to others. In the example below, the first molecule’s pharmacophore constellations fall predominantly in the clusters statistically linked to target A. This molecule should therefore be ranked high for target A assays. On the other hand, the second molecule’s pharmacophore constellations are mixed from different target constellations. This molecule would therefore be ranked lower.

workflow2

The JPS can be viewed as a very high-dimensional space of pharmacophore constellations that contains small subspaces that can be labeled with the corresponding property or target associated with molecules falling into them.

jps_example
The picture on the left shows an example JPS. Each colored ellipsoid indicates the subspace within the JPS that contains pharmacophore constellations relevant to the properties shown. As seen in this example, the JPS can mix target specific properties (such as RXR agonistic binding) with general drug properties (such as blood-brain barrier permeability). This way, the JPS can consider all properties at once and prioritize the compounds highest that have the desired combination of properties (such as RXR binder, hERG inactive, and BBB permeable) while avoiding undesired properties (such as binding to RAR) which could lead to side-effects.

This is a very powerful way to reach a set of high-quality druglike lead candidates in a very short time.

Highest Quality Lead Candidates

jps_retro
The underlying algorithms of the JPS approach have been thoroughly tested and fine-tuned over the years. In every single retrospective study performed, JPS outperformed all competing approaches by a wide margin. An example study result is shown in the table on the right. This study compared the four leading in-silico modeling and screening approaches to JPS for the NCI cancer cell-based assay using the BEDROC score.

It shows that JPS offers consistently high accuracy across many modes-of-action, even if multiple targets may be involved in an observed activity. Further analysis shows high complementarity compared to other approaches (which translates into a high likelihood for novel leads).

In summary, Acelot’s drug leads have a much higher likelihood of success and a much lower risk of failure than leads discovered by competing technologies.

Insights into pharmacophore constellations

insights
Since the labeled subspaces in the JPS are associated with the statistically relevant 3D constellations of
pharmacophoric points, JPS can be used to gain insights into binding mechanisms involved in an observed activity. No target structure information is needed for this purpose. Solely the positives and negatives from chemical or biological assays are sufficient. As shown in the JPS visualization on the left, the JPS algorithm will output the most significant pharmacophore constellations, together with their approximate relative distances.

The JPS approach even works for cell-based/phenotypic assays that may exhibit multiple target interactions. In this case, JPS can be used to identify larger “meta”-clusters within the high-dimensional space, one for each target involved. This way, cell-based assays can be deconvoluted and at the same time, other compounds with similar multi-target interactions (i.e., similar biological properties) can be identified.

JPS in the Drug Discovery Pipeline

jps_cycle
Acelot’s JPS approach works in concert with biological and biochemical assays as shown on the right. Without any additional knowledge of the ligand or target structure, JPS can automatically learn pharmacophore constellations that matter for binding and/or some activity measured in an assay. A very small number of assay results (in the 10s of measurements) is sufficient as input.

Once the JPS is built, it can be used to rapidly screen millions of compounds for additional potential hits based on one or more targets and/or ADME/Toxicity properties. These additional compounds can then be assayed again and the result can be fed back into JPS to further improve the prediction model. This iterative refinement combined with the small number of input molecules needed, can lead to a significant cost and time reduction in screening, without loss of lead quality.

Acelot’s Graph-Theory Based Tools

Besides the 3D pharmacophore analysis method, Acelot has developed numerous “2D” drug discovery tools based on graph theory. These tools analyze the scaffolds of molecules in order to find other compounds with similar (but not too similar) structures, to discover fragments that are statistically over- or under-represented among a set of compounds, or to predict drug-relevant properties such as blood-brain barrier permeability or hERG activity.

In contrast to other “1D” or “2D” drug discovery software tools that rely on fingerprints or other descriptors, Acelot’s tools provide a higher prediction accuracy, and – more importantly – complementary results. Complementarity is important in order to discover truly novel treatments.