Computer implemented method and system for small molecule drug discovery
In a small molecule drug discovery method, a transition state for a specific enzyme is modelled using quantum mechanics and molecular dynamics based simulation of the enzyme and substrate reaction; data defining the transition state (a ‘quantum pharmacophore’) is fed to a machine learning engine configured to generate transition state analogues, such as enzyme inhibitors.
1. A small molecule drug discovery method comprising the steps of:
(a) generating data representing a transition state occurring within a specific enzyme and a substrate reaction, using quantum mechanics and molecular dynamics based simulation of the specific enzyme and substrate reaction, running on a computer-implemented simulation engine, in which the data representing the transition state further includes a description of the quantum chemical properties of the specific enzyme and the substrate in the transition state;
(b) storing the data representing the transition state (‘the quantum pharmacophore’) in a memory;
(c) passing the quantum pharmacophore into a computer-implemented machine learning engine configured to generate data defining transition state analogues, in which transition state analogues are small molecules that enable the specific enzyme to enter its transition state; and
in which the computer-implemented machine learning engine is a generative ML system configured to optimise a cost function; and
in which the cost function reflects a similarity of each generated transition state analogue to the quantum pharmacophore, in which the similarity is assessed in terms of one or more of the following molecular properties: distribution of charges or Van der Waals dispersion or tensor network like description of the transition state or properties assessed via wetlab assays and feedback loops.
2. The method of claim 1 in which the simulation engine generates data and information that allows for the development of descriptions of the transition state that form the quantum pharmacophore.
3. The method of claim 1 in which the quantum mechanics and molecular dynamics based simulation of the enzyme/substrate reaction uses a hybrid quantum mechanics (QM) and classical model.
4. The method of claim 1 in which the quantum mechanics and molecular dynamics based simulation of the enzyme/substrate reaction combines or uses a combination of a pure QM model and a (quantum mechanics/molecular mechanics) QM/MM model.
5. The method of claim 1 in which the quantum mechanics and molecular dynamics based simulation of the enzyme/substrate reaction uses one or more of the following: perturbative corrections, extensions of a molecular mechanics framework, or uses extensions of a molecular mechanics framework using the computer implemented machine learning on QM simulation data.
6. The method of claim 1 in which the quantum mechanics and molecular dynamics based simulation of the enzyme/substrate reaction only requires a crystal structure of the specific enzyme and knowledge of the specific enzyme/substrate reaction.
7. The method of claim 1 in which the quantum mechanics and molecular dynamics based simulation of the enzyme/substrate reaction only requires a crystal structure of the specific enzyme and knowledge of one or more of the following factors of the specific enzyme/substrate reaction: (i) input substrate(s) and co-factors including non-enzyme chemicals that impact the reaction; (ii) output product(s) of the reaction; (iii) any allosteric binders required for the specific enzyme to be catalytically competent.
8. The method of claim 1 in which the quantum pharmacophore defines quantum mechanics data that is relevant to a binding between the specific enzyme and substrate.
9. The method of claim 1 in which the quantum pharmacophore is a description of the structural and/or quantum chemical properties of the specific enzyme and substrate in the transition state.
10. The method of claim 9 in which the quantum chemical properties of the specific enzyme and substrate in the transition state captures a quantum entanglement properties of the transition state including non-local interactions and/or features that are calculated from a tensor network based description of the transition state.
11. The method of claim 1 in which the quantum pharmacophore captures an arrangement of atoms, groups, or functionalities in a small molecule required for specific interactions with its biological target and its activity.
12. The method of claim 1 in which the quantum pharmacophore is a template for representing properties of a compound that would replicate a binding mode and affinity of the substrate in the transition state to the enzyme.
13. The method of claim 1 in which the quantum pharmacophore is based on substrate features, constraints placed on a ligand by an enzyme configuration or a combination of both.
14. The method of claim 1 in which the computer implemented machine learning engine is a generative ML system configured to optimise a cost function that includes constraints based on medicinal chemistry input.
15. The method of claim 14 in which the medicinal chemistry input relates to an avoidance of specific chemical moieties, or divergence from the quantum pharmacophore by constraining properties.
16. The method of claim 1 in which the computer implemented machine learning engine uses the quantum pharmacophore to guide generative models to areas of chemical space that mimic the transition state and hence can inherit the properties of the transition state, including a binding mode.
17. The method of claim 1 in which the computer implemented machine learning engine identifies and discards transition state analogues which are not drug-like or have predefined properties.
18. The method of claim 1 in which outputs from the computer implemented machine learning engine are assessed via wetlab assays and feedback loops to enable the identification of candidates to provide designs of potential inhibitors.
19. The method of claim 1 in which feedback loops include predictive models for predefined properties.
20. The method of claim 1 in which intermediate wet lab results are fed back both into simulation and ML models to constantly improve the search and ultimately find a lead series.
21. The method of claim 1 in which in silico and lab-based work is done on the computer implemented machine learning engine outputs to filter out the transition state analogues which are not drug-like, have predefined properties including toxicity above a certain threshold or similarity to the substrate or reactant above a certain threshold, in which similarity is assessed in terms of one or more of the following molecular properties: distribution of charges or Van der Waals dispersion.
22. The method of claim 1 in which in silico and lab-based work is done on the computer implemented machine learning engine outputs to identify which transition state analogues have a binding to the specific enzyme that is above a certain threshold.
23. The method of claim 1 in which the generated transition state analogues are refined to trade off binding strength with other properties which impact drug efficacy using another set of ML models.
24. The method of claim 1 in which the transition state analogues are de novo compounds.
25. The method of claim 1 in which the transition state analogue is an enzyme inhibitor, in which the enzyme inhibitor is a kinases, protease or metalloenzyme inhibitor.
26. The method of claim 1 in which the transition state analogue is a crop protection compound.
27. A small molecule drug virtual screening method comprising the steps of:
(a) generating data representing a transition state occurring within a specific enzyme and a substrate reacting, using quantum mechanics based simulation of the specific enzyme and substrate reaction, running on a computer-implemented simulation engine, in which the data representing the transition state further includes a description of quantum chemical properties of the specific enzyme and the substrate in the transition state;
(b) storing the data representing the transition state (‘the quantum pharmacophore’) in a memory;
(c) passing the quantum pharmacophore into a computer-implemented machine learning engine configured to virtually screen known compounds to identify potential candidates as transition state analogues, in which transition state analogues are small molecules that enable the specific enzyme to enter its transition state;
and in which the computer implemented machine learning engine is a generative ML system configured to optimise a cost function; and
in which the cost function reflects a similarity of each generated transition state analogue to the quantum pharmacophore, in which similarity is accessed in terms of one or more of the following molecular properties: distribution of charges or Van der Waals dispersion or tensor network like description of the transition state or properties assessed via wetlab assays and feedback loops.