LeadInvent - Technology Page


Contact details

Plot No. 32, Sector 5, IMT Manesar, Gurgaon, Haryana 122052, India

Email: contact@leadinvent.com

View Larger Map

LeadInvent has developed unique molecule simulation platform (christened LIDiscoverEngine). This simulation platform enables scientist to design and study molecules virtually on supercomputing platform before such molecules are synthesized and tested experimentally. This capability of virtual design and testing saves considerable experimental cost and reduces discovery time. More over LIDiscoverEngine provides atomic level insight and unravels the energetics of binding for a protein ligand system. This unique capability guides scientist to discover drugs through a rational approach rather than conventional hit and trial.

Accurately predict molecule binding

Understanding atomic level contacts, define essential pharmacophore , carve out unoccupied space within active site, suggest changes on molecule without affecting binding.

Estimating accurate binding between a target protein and small molecule is pivotal.
Built upon iterative experimental feedback, LeadInvent has developed unique simulation methodology and mathematical grading (scoring) functions that accurately docks and predict binding of a molecule to its target protein.

Calculate accurate binding energies between ligand and protein

Predict experimental IC50 Prioritize synthesis of molecules

Accurate binding energy scoring is central to a simulation that predicts protein ligand interaction. However most current simulation methods rely on mathematical approximations to generate such scores.

LeadInvent’s binding energy scoring platform is a unique approach. Our method is at the interface of quantum calculation and molecular mechanics. This has enabled us to estimate binding energies with unmatched accuracies and strong correlation with experimental IC50’s.

The graph here depicts accuracy of a leading simulation software in comparison to LeadInvent’s simulation package (LIDiscoverEngine). Each dot represents a molecule and its relative energy value predicted by respective simulation software. X axis represents experimental IC50 values for each molecule and Y axis represents the software predicted values.

It is quite evident that the prediction from the leading software is scattered with no correlation with experimental values . Where as results from LIDiscoverEngine are highly grouped. Because of such strong correlation between LIDiscoverEngine and actual experimental results. LeadInvent can upfront estimate potency of molecules even before they are chemically synthesized and tested. The dot highlighted with the red circle in the LeadInvent data graph was also successfully predicted by LIDiscoverEngine as a molecule with poor conformation to target protein.

Capture dynamic binding

Going beyond crystal 3D Unraveling Protein Conformations

Proteins are complex three dimensional bio molecule made up of hundreds of thousands of atoms. Typically a 3D structure of protein could be understood from crystallography technique. While this technique provides very useful information about a protein and its over all 3D shape. It typically represents one of the many low energy conformation available to a protein. Protein molecules are flexible bio molecules and subtle changes in their conformation dictate major functional roles for them. Hence while working with proteins as drug targets. It is important to have tools and capacities to understand and work with different conformations of a target protein from a drug design perspective. LeadInvent has developed extensive expertise and technology to simulate and handle various conformations sampled by a target protein. This capability of LeadInvent has been used extensively across multiple discovery programs executed by the company and has unraveled key understanding in such programs.

Case Study: LeadInvent worked on a target protein for which the data is represented above. According to experimental results. The mutated protein has a 20X higher activity for its substrate than the native protein. However the reason for such higher activity was unexplained. The graph here depicts different conformations of this target protein (average RMSD) . The red graph represents different conformations of the protein carrying the single amino acid mutation (mutated protein). Where as the black graph depicts native protein without mutation. The two protein conformations are shown in the panel to the right. It is clear that the mutated protein attains a specific conformation that opens up the protein structure giving early clue to the 20X higher activity of the protein for its substrate. This case study highlights the necessity for deconvulating protein conformations and establishes the capabilities of the team to use such information in its drug discovery endeavors

Understanding bound vs unbound states of molecules

Small molecules are generally flexible molecules with certain degrees of freedom around their rotatable bonds. Because of this inherent structural freedom a molecule could exist in multiple shapes (conformations). When molecules are free (not bounded to their target proteins) they are able to sample multiple different confirmations. Where as when molecules are bound to their target proteins. They are restricted by the target protein and hence molecules exist in limited conformations.

LIDiscoverEngine provides key insight into such studies and guides scientist in designing conformationally stable molecules.

Here two molecules were designed and tested for activity against a target protein . The two molecules were almost identical except a single atom change at position A1 and A2. Compound with A1 change reported and experimental IC50 of approximately 50 nM where as compound with A2 change was found to have IC50 greater than 1 uM. In the right image panel unbound molecules are depicted by red ice sickle where as bound molecule conformations are depicted by green ice sickle.

It is clear that molecule with A2 change has almost 50% of the unbound molecule prearranged in the conformations that overlaps with the protein bound state. Where as molecule with A1 change has less than 10% free molecules conformations that over lap with bound conformations.

This clearly points out the subtle conformational energetic penalties paid by molecules as they go form an unbound state to a bound state and dictate their activity profile against a protein target.

Red well represent unbound molecule conformations. Green well represent the protein bound conformations of the same molecule

Predict activity of a molecule across 8000 targets before undertaking the first experiment

Understand what other targets could be engaged by a molecule, Drive better selectivity for molecules, repositioning molecule from one target to another.

One of the biggest challenge in drug discovery is to design and develop molecules that selectively show activity against a particular target protein. A possible solution to this is to test activity of molecule across multiple targets for possible cross target activity. While experiments take several weeks and cost tens of thousands of dollars for testing cross target activity of molecules. LIDiscoverEngine is able to virtually simulate and test activity of a molecule across 8000 proteins within a single day. Built on top of 1 million literature reported bio activity data points for small molecule. This module of LIDiscoverEngine has been extensively tested.

Case Study: Predicting off target activity of kinase inhibitors.

Kinases are important proteins that act like molecular switches regulating important cellular functions making them important drug targets. However kinases are very similar to each other and share a common drug binding site (ATP binding site). Because of high similarity among different kinases it is known that kinase inhibiting molecules show activity across multiple kinases. LeadInvent team tested over 14 different kinase compounds and successfully predicted activity of all 14 molecules across 400 + kinases with an accuracy of ~ 80% when compared to experimental data.

A red dot depicts a kinase for which the test molecule shows experimental activity less than 10 uM.

A Blue dot depicts a kinase for which LIDiscoverEngine predicts activity of a molecule.

Whenever simulation technique predicts activity of a molecule against a kinase as predicted by experimental studies. Blue dot over lap the red dot. Hence each overlap of red and blue dot confirms computer prediction of experimental result.

Case study CHIR265 : The above image depicts a kinome tree.

A Red dot depicts a kinase against which CHIR265 has been experimentally reported to be active. The blue dot depicts a kinase against which LIDiscoverEngine has predicted activity of CHIR265. It is clear ( from the overlap of blue and red dots) that LIDiscoverEngine is able to predict activity of Chir265 across multiple kinases with above 80% accuracy

Screen multi million compound library

Harnessing power of supercomputing for virtually testing molecules

LIDiscoverEngine has built in module to screen large number of virtual molecules against a target protein. Implemented on a supercomputing architecture this technology harnesses the power of parallel computing along with seamless connectivity across LeadInvent’s multiple simulation modules such as automated ligand preparation, protein conformation generators, parallel docking, and scoring among others.

LeadInvent currently maintains a virtual library of over 8 million compounds that are screened against a target for finding early hit compounds, define pharmacophore of binding and predicting chemical space suitable for a given proteins active site.

This technology focuses scientist on testing only the promising molecules saving significant cost and efforts as compared to hit and trial screening.

This module was used by LeadInvent for a Boston based cancer research company and resulted in 24 hit compounds that were reported positive in experimental studies.

Building molecule inside the protein cavity using small molecule fragments

LIDiscoverEngine combines the power of building novel molecules by attaching small molecule fragments along with unmatched parallel processing algorithms. Building block by block from over 2500 small molecule fragments. This technology build molecules in an automated fashion from within the active site.