System and method for automated retrosynthesis
A system and method for automated retrosynthesis which can reliably identify valid and practical precursors and reaction pathways. The methodology involves a k-beam recursive process wherein at each stage of recursion, retrosynthesis is performed using a library of molecule disconnection rules to identify possible precursor sets, validation of the top k precursor sets is performed using a transformer-based forward reaction prediction scoring system, the best candidate of the top k precursor sets is selected, and a database is searched to determine whether the precursors are commercially available. The recursion process is repeated until a valid chain of chemical reactions is found wherein all precursors necessary to synthesize the target molecule are found to be commercially available.
1. A system for recursive retrosynthesis of molecules, comprising:
a computing device comprising a memory, a processor, and a non-volatile data storage device;
a data curation platform, comprising a first plurality of programming instructions stored in the memory which, when operating on the processor, causes the computing device to:
scrape data from a plurality of data sources accessible via the internet;
parse the scraped data into a format that may be stored in a relational database;
construct a knowledge graph from the information in the scraped data, the knowledge graph representing informational entities as vertices of a graph and relationships between the entities as edges of the graph;
a machine learning algorithm operating on the computing device, the machine learning algorithm having been trained to predict valid chemical reactions;
a commercial availability database stored on the non-volatile data storage device, the commercial availability database comprising a list of commercially available molecules;
a disconnection rules library stored on the non-volatile data storage device, the disconnection rules library comprising a set of rules for breaking of molecules into precursors based on the knowledge graph;
a recursion engine comprising a second plurality of programming instructions stored in the memory which, when operating on die processor, causes the computing device to:
track stages of recursive retrosynthesis;
for each stage of recursive retrosynthesis:
select a molecule for retrosynthesis;
send chemical notation for the molecule to a retrosynthesis engine;
receive a pre-determined number of sets of possible precursors for die molecule;
create a hypernode of a hypergraph for each of the sets of possible precursors;
perform forward reaction prediction of the possible precursors in each hypernode using the machine learning algorithm;
assign a reinforcement learning reward to the hypernode based on the forward reaction prediction;
backpropagate the reinforcement learning reward through previous stages of the recursive retrosynthesis;
search the commercial availability database to determine whether the precursors are commercially available;
end die recursive retrosynthesis when a set of precursors has been found which is commercially available; and
a retrosynthesis engine comprising a third plurality of programming instructions stored in the memory which, when operating on the processor, causes the computing device to:
receive the chemical notation for the molecule from the recursion engine;
using the disconnection rules library, generate the pre-determined number of sets of possible precursors for the molecule; and
send the sets of possible precursors back to the recursion engine.
2. The system of claim 1 , further comprising a forward reaction scoring function which assigns a score to the forward reaction prediction.
3. The system of claim 2 , wherein the forward reaction scoring function comprises an upper confidence bound (UCB) score.
4. The system of claim 2 , wherein the forward reaction scoring function utilizes a benchmarking tool comprising a set of retrosynthesis steps that are known to be valid.
5. The system of claim 1 , further comprising a confidence level assessor comprising a fourth plurality of programming instructions stored in the memory which, when operating on the processor, causes the computing device to:
for each stage of recursion, receive a top-k number of sets of possible precursors of the molecule, the top-k number being greater than the predetermined number;
compare each of the top-k number of sets against a confidence threshold;
eliminate sets of top-k number of sets that fall below the confidence threshold until the remaining top-k number of sets equals the predetermined number.
6. A method for recursive retrosynthesis of molecules, comprising die steps of:
using a data curation platform operating on a computing device:
scraping data from a plurality of data sources accessible via the internet;
parsing the scraped data into a format that may be stored in a relational database;
constructing a knowledge graph from the information in the scraped data, the knowledge graph representing informational entities as vertices of a graph and relationships between the entities as edges of the graph;
using a recursion engine operating on the computing device:
tracking stages of recursive retrosynthesis;
for each stage of recursive retrosynthesis:
selecting a molecule for retrosynthesis;
sending chemical notation for the molecule to a retrosynthesis engine operating on the computing device;
receiving a pre-determined number of sets of possible precursors for the molecule from a retrosynthesis engine operating on the computing device;
creating a hypernode of a hypergraph for each of the sets of possible precursors;
performing forward reaction prediction of the possible precursors in each hypernode using a machine learning algorithm operating on the computing device, the machine learning algorithm having been trained to predict valid chemical reactions;
assigning a reinforcement learning reward to the hypernode based on the forward reaction prediction;
backpropagating the reinforcement learning reward through previous stages of the recursive retrosynthesis;
searching a commercial availability database to determine whether the precursors are commercially available, the commercial availability database comprising a list of commercially available molecules and being stored in a non-volatile data storage device on the computing device;
ending the recursive retrosynthesis when a set of precursors has been found which is commercially available; and
using a retrosynthesis engine operating on the computing device:
receiving the chemical notation for the molecule from the recursion engine;
generating the pre-determined number of sets of possible precursors for the molecule using a disconnection rules library stored on a non-volatile data storage device, the disconnection rules library comprising a set of rules for breaking of molecules into precursors based on the knowledge graph; and
sending the sets of possible precursors back to the recursion engine.
7. The method of claim 6 , further comprising the step of using a forward reaction scoring function to assign a score to the forward reaction prediction.
8. The method of claim 7 , wherein the forward reaction scoring function comprises an upper confidence bound (UCB) score.
9. The method of claim 7 , wherein the forward reaction scoring function utilizes a benchmarking tool comprising a set of retrosynthesis steps that are known to be valid.
10. The method of claim 6 , further comprising the step of using a confidence level assessor operating on the computing device to:
for each stage of recursion, receive a top-k number of sets of possible precursors of the molecule, the top-k number being greater than the predetermined number;
compare each of the top-k number of sets against a confidence threshold;
eliminate sets of top-k number of sets that fall below the confidence threshold until the remaining top-k number of sets equals the predetermined number.