System and method for optimizing chemical reactions using machine learning

Abstract

The present disclosure provides methods and systems for optimizing chemical reactions through machine learning. A chemical space is defined by grouping the potential chemicals based on the selected characteristics. Representative chemicals are selected from each group to assemble a test kit. The test kit is then used to identify the optimal catalyst or ligand for the catalytic chemical reaction. In some embodiments, the system may recommend a test kit, receive results from an experiment, generate a distance matrix, and rank potential chemicals based on scores obtained from their representative chemicals. The method involves reducing the dimensionality and normalization distance of the chemical feature. The disclosed embodiments may suggest potential chemicals for optimizing chemical reactions by ranking representative chemicals based on scores obtained from them.

Description

BACKGROUND OF THE DISCLOSURE

[0001] Related fields

[0002] The present disclosure relates to optimizing chemical reactions. In particular, the present disclosure relates to a system and method for optimizing catalytic chemical reactions using a test kit and machine learning, such as by using design of experimental methods to optimize the selection of one or more catalysts and / or ligands.

[0003] Description of related technologies

[0004] Catalysis is a process of adding a non-consumable material to a chemical reaction to increase the rate, efficiency, or otherwise modify the chemical reaction to achieve desired process parameters or results. A catalyst is a substance that accelerates a chemical reaction and / or reduces the activation energy, temperature, or pressure required to initiate a chemical reaction. A catalyst is not consumed in the reaction and generally remains unchanged after the chemical reaction is complete. A small amount of catalyst is usually sufficient to facilitate a chemical reaction.

[0005] Approximately 90% of all commercially produced chemical products involve catalysts at some stage in their production process. Chemists in academia and industry worldwide often optimize catalytic reactions by changing catalytic materials, reactants, solvents, or reaction conditions to increase yield, efficiency, or cost-effectiveness.

[0006] In a process chemistry laboratory, to be used on a large scale in a manufacturing plant, research chemists or laboratory technicians design a synthetic route by conducting several test experiments with approximately 20 catalysts to optimize a single catalytic chemical reaction. Finding a suitable catalyst can take weeks, resulting in high costs and additional delays in bringing the product to market. In most cases, the catalyst is selected to increase the reaction yield and suppress unwanted side reactions while being safe, cost-effective, and environmentally friendly.

[0007] Currently, the selection of catalysts depends largely on human intuition and is reduced to the most common materials in the laboratory or chemical supplier inventory, thus overlooking a large number of other materials within the chemical space, some of which may be more effective than traditional choices. Therefore, only a fraction of the possible catalysts can usually be screened, which may not necessarily lead to the best choice.

[0008] Therefore, in order to follow and improve such the state of the art, it is desirable to find a new method to select the best catalyst, ligand, or other chemical from a large number of chemical catalogs without having to conduct experimental trials on each chemical.

[0009] SUMMARY

[0010] According to the present disclosure, a method for assembling, by a computer, a test suite for optimizing catalytic chemical reactions using machine learning is disclosed. The method may include the following acts: parameterizing, by a computer, a catalyst for a catalytic chemical reaction for corresponding chemical features specific to the catalytic chemical reaction; grouping the parameterized catalysts into a given number of clusters based on chemical featurization and molecular descriptors of the catalysts, the clusters covering the entire chemical space of the catalysts; selecting, by a computer, a representative catalyst from each cluster according to a specific given criterion; and assembling the test suite with the selected representative catalysts as included components.

[0011] This method is different from the known prior art in that, inter alia:

[0012] 1. Incorporating the physical test suite into the workflow with the aim of identifying optimized ligands and catalysts without knowledge of the reaction.

[0013] 2. The test suite can optionally be standardized to support multiple reactions.

[0014] 3. The clustering algorithm includes identifying an optimal set of catalysts or ligands for the test suite by combining chemical features and commercial viability and / or availability.

[0015] Advantageous and thus exemplary further improvements of the present disclosure emerge from the relevant dependent claims, the description and the associated drawings.

[0016] One of those exemplary further embodiments of the disclosed method includes determining the corresponding chemical features by chemical informatics and computational modeling on a computer.

[0017] Another of those exemplary further embodiments of the disclosed method includes grouping implemented by a computer through k-means clustering, density-based spatial clustering of applications with noise (“DBSCAN”), spectral clustering, Gaussian mixture models or other clustering algorithms known to those of ordinary skill in the relevant art. For example, density-based, distribution-based, centroid-based or hierarchical clustering algorithms can be used.

[0018] Another exemplary embodiment of the present disclosure selects chemicals based on a combination of chemical features, commercial viability and / or source availability of the catalysts or ligands.

[0019] Another exemplary embodiment of the disclosed method includes storing all available catalysts or ligands in a database connected to a computer.

[0020] Another exemplary embodiment uses a test kit having a specific given number of catalyst or ligand components, the test kit assembled by using the methods disclosed herein.

[0021] Another embodiment of the present disclosure includes a method for optimizing a catalytic chemical reaction using a computer - supported test kit, the method including the following optional acts: performing a standardized experiment on the catalytic chemical reaction using components in the test kit; inputting the result data of the performed experiment into a computer; using a machine - learning algorithm (such as but not limited to clustering algorithms, regression algorithms, classification algorithms, unsupervised machine - learning algorithms, regression models, etc.) running on the computer to interpolate between a given number of clusters in the chemical space covered by all available catalysts or ligands; using a machine - learning regression model to predict the best - fitting catalyst for the catalytic chemical reaction in the interpolated parameter space of all available catalysts or ligands; and performing the catalytic chemical reaction with the predicted catalyst or ligand.

[0022] An exemplary further embodiment of this disclosed method includes providing a network interface via a computer, through which a user can upload the yield values of the chemical reactions from the performed experiments.

[0023] It is to be understood that all aspects of those preferred further improvements can be combined together, even if not explicitly stated, unless it is clearly impossible due to the nature of the corresponding features.

[0024] A system including one or more computers can be configured to perform specific operations or acts by installing software, firmware, hardware, or a combination thereof. Such software, firmware, or hardware can cause the system to perform the desired actions when running. Additionally, one or more computer programs can be designed to perform specific operations or acts by including instructions that, when executed by a data - processing device, cause the device to perform the desired behavior.

[0025] In one general aspect, the method may include selecting a chemical space for grouping. Each chemical space may be defined by a plurality of prospective chemicals. The method may also include selecting a plurality of chemical features, where each chemical feature corresponds to a plurality of prospective chemicals. Moreover, the method may include grouping the plurality of prospective chemicals within a grouping space based on the plurality of chemical features. Additionally, the method may include selecting a plurality of representative chemicals from the prospective chemicals. Each representative chemical corresponds to a group of the plurality of prospective chemicals grouped within the grouping space. Further, the method may include assembling a test set having a plurality of test chemicals. Each test chemical corresponds to one of the plurality of representative chemicals. Other embodiments of this aspect include corresponding computer systems, apparatuses, and computer programs recorded on one or more computer storage devices. Each may be configured to perform the operations of the method.

[0026] In one general aspect, the method may include embodiments that may include one or more of the following features as part of the method. The method may include where the grouping space is defined by a plurality of chemical features. The grouping space may be a dimensionally reduced space of the plurality of chemical features. The method may also include the act of generating a plurality of reduced chemical spaces, where each reduced chemical space is a dimensionally reduced space of the chemical space. The method may include calculating a plurality of distances, where each distance of the plurality of distances is a distance between one of the plurality of prospective chemicals and one of the plurality of test chemicals. The plurality of test chemicals may be a subset of the plurality of prospective chemicals. The plurality of prospective chemicals may not include the plurality of test chemicals. The method may include calculating a plurality of distance metrics, where each distance metric of the plurality of distance metrics is one minus the distance between one of the plurality of prospective chemicals and one of the plurality of test chemicals.

[0027] In one general aspect, the method may include embodiments that may include one or more of the following features as part of the method. The method may include averaging each of a plurality of distance metrics in an all-reduced chemical space to generate a plurality of average distance metrics. The method may include generating a plurality of weights, where each of the plurality of weights corresponds to one of a plurality of outcomes, and where each of the plurality of outcomes corresponds to one of a plurality of test chemicals. The plurality of weights may be determined according to an exponential function. Each of the plurality of distances of the distance metrics may be multiplied by a corresponding one of the plurality of weights. A maximum value of each corresponding potential chemical may be taken among all of the multiplied plurality of distance metrics between the respective potential chemicals and the plurality of test chemicals. The method may include sorting the maximum values of each corresponding potential chemical. The method may include testing the plurality of test chemicals to determine a plurality of outcomes, where each of the plurality of outcomes corresponds to a respective outcome of one of the plurality of test chemicals.

[0028] In one general aspect, the method may include embodiments that may include one or more of the following features as part of the method. A plurality of outcomes may be uploaded to a server. The method may include using a plurality of predictive features to select a prediction space to provide a prediction to an outcome space, where the plurality of outcomes define data points within the prediction space that map to the outcome space. The outcome space may include chemical yield. The outcome space may be a byproduct metric. The outcome space may be a single parameter. The prediction space may include a grouping space. The method may include selecting a chemical from among a plurality of potential chemicals corresponding to an optimized value of the outcome space. The prediction space may, in some of its aspects, include a plurality of chemical features therein. A computer may be configured to use regression fitting to fit the plurality of outcomes to map the plurality of predictive features to the outcome space. A computer may be configured to use the plurality of outcomes to train a learner.

[0029] In one general aspect, the method may include embodiments that may include one or more of the following features as part of the method. The learner may include at least one of an artificial neural network, k-nearest neighbor, decision tree, random forest, support vector machine, Bayesian regressor, and / or ensemble. The plurality of predictive features may include at least one of a space-filling feature, a bulk feature, an orientation feature, an electrical property feature, an input voltage (Vin), frontier molecular orbitals (frontier Mos), Fukui function, NBO analysis, NMR tensor, steric property, Sterimol L, B1, B5, B1, B5, quadrant analysis, octant analysis, total volume, buried volume, dipole moment, solvation energy, dispersion of potential, etc. The plurality of chemical features may include a plurality of chemical characterizations. The plurality of chemical features may include a plurality of molecular descriptors. The chemical space may include a plurality of catalysts. The chemical space may include a plurality of ligands. The plurality of chemical features may be determined by at least one of chemoinformatics and computational modeling on a computer. The plurality of chemical features may be stored in a database executed by a computer.

[0030] In one general aspect, the method may include embodiments that may include one or more of the following features as part of the method. The method may include testing a plurality of test chemicals to determine a plurality of results, where each of the plurality of results corresponds to a respective result of one of the plurality of test chemicals; selecting a prediction space using the plurality of predictive features to provide a prediction to an outcome space, where the plurality of results define data points within the prediction space; using regression to map the plurality of predictive features to the outcome space; determining, based on the prediction, the best catalyst or best ligand among all available catalysts or ligands for catalyzing a chemical reaction in the prediction space; and / or catalyzing the chemical reaction with the best catalyst or best ligand based on the prediction.

[0031] In one general aspect, the method can include embodiments that can include one or more of the following features as part of the method. Multiple results can be uploaded to a server. The method can include using multiple predictive features to select a prediction space to provide predictions to a result space, where the multiple results define data points within the prediction space. A computer can be configured to map the multiple predictive features to the result space using regression by fitting the multiple results to the predictive features and the result space. A grouped space can be generated by performing dimensionality reduction on multiple chemical features. The dimensionality reduction can be implemented on a computer using principal component analysis. The chemical space can be defined by multiple phosphine ligands for cross-coupling reactions. The cross-coupling reaction can include one of Suzuki catalysis and Buchwald catalysis. The multiple test chemicals can include 24 chemicals. The multiple chemical features can include at least one of space-filling features, bulk features, orientation features, electrical features, input voltage (Vin), frontier molecular orbitals, Fukui functions, NBO analysis, NMR tensors, steric properties, Sterimol L, B1, B5, B1, B5, quadrant analysis, octant analysis, total volume, buried volume, dipole moment, solvation energy, and / or potential energy dispersion. Embodiments of the techniques can include hardware, methods or processes, or computer-readable media.

[0032] In one general aspect, a method can include assembling a test set having multiple test chemicals, where each test chemical corresponds to one representative chemical among multiple representative chemicals; testing the multiple test chemicals to determine multiple results, where each of the multiple results corresponds to a respective result of a respective one of the multiple test chemicals; and determining, based on predictions, the best catalyst or best ligand among all available catalysts or ligands for catalyzing a chemical reaction in a prediction space. Other embodiments of this aspect include corresponding computer systems, devices, and computer programs recorded on one or more computer storage devices, each configured to perform the operations of the method.

[0033] Embodiments can include one or more of the following features. The method can include: generating multiple reduced chemical spaces, where each reduced chemical space is a space with reduced dimensionality of the chemical space; calculating multiple distances, where each of the multiple distances is a distance between one potential chemical among multiple potential chemicals and one test chemical among the multiple test chemicals; calculating multiple distance metrics, where each of the multiple distance metrics is one minus the distance between one of the multiple potential chemicals and one of the multiple test chemicals; and / or averaging each of the multiple distance metrics across all of the reduced chemical spaces to generate multiple average distance metrics.

[0034] In one general aspect, the method may include embodiments that may include one or more of the following features as part of the method. The method may include generating a plurality of weights, where each weight in the plurality of weights corresponds to one of a plurality of outcomes, and where each outcome in the plurality of outcomes corresponds to one of a plurality of test chemicals. The plurality of weights may be determined according to an exponential function. Each of a plurality of distances of a distance metric may be multiplied by a corresponding weight in the plurality of weights. A maximum value of each corresponding potential chemical may be taken across all of the plurality of distance metrics that may be multiplied between the corresponding potential chemical and the plurality of test chemicals.

[0035] In one general aspect, the method may include embodiments that may include one or more of the following features as part of the method. The method may include ranking the maximum values of each corresponding potential chemical to find the best catalyst or best ligand; using a plurality of predictive features to select a prediction space to provide a prediction to an outcome space, where the plurality of outcomes define data points within the prediction space; and / or using regression to fit data points within the plurality of predictive features to the outcome space. The plurality of predictive features may include at least one of space-filling features, bulk features, orientation features, electronic features, input voltage (Vin), frontier molecular orbitals, Fukui functions, NBO analysis, NMR tensors, steric properties, Sterimol L, B1, B5, B1, B5, quadrant analysis, octant analysis, total volume, buried volume, dipole moment, solvation energy, and / or potential energy dispersity, etc. The method may include catalyzing a chemical reaction with the best catalyst or best ligand according to the prediction. Embodiments of the techniques may include hardware, methods or processes, or computer-readable tangible media.

[0036] In one general aspect, the method may include parameterizing, by computer, a catalyst or ligand for a corresponding catalytic chemical reaction with respect to chemical characteristics specific to the catalytic chemical reaction. The method may further include grouping the parameterized catalysts or ligands based on their chemical characteristics into a given number of clusters that cover the chemical space of the catalysts or ligands. Additionally, the method may include using a computer to select a representative catalyst or ligand from each cluster according to a predetermined criterion. Additionally, the method may include assembling a test set with the selected representative catalysts or ligands as components. Other embodiments of this aspect include corresponding computer systems, devices, and computer programs recorded on one or more computer storage devices, each configured to perform the operations of the method.

[0037] Embodiments may include one or more of the following features. Corresponding chemical features can be determined through chemoinformatics and computational modeling on a computer. The method can use a computer to perform a grouping operation by using k-means clustering. The selection can be based on a combination of chemical features and the commercial viability and / or source availability of catalysts or ligands as a predetermined criterion. All available catalysts and / or ligands can be stored in a database connected to the computer.

[0038] In one general aspect, the method may include embodiments that may include one or more of the following features as part of the method. The method may include: performing a standardized experiment on a catalytic chemical reaction with components in a test suite; inputting the result data of the performed experiment into a computer; using a machine learning regression model running on the computer to interpolate between a given number of clusters in the chemical space covered by all available catalysts or ligands; using the machine learning regression model to predict the best-fitting catalyst for a catalytic chemical reaction in the chemical space covered by all available catalysts or ligands; and / or performing a catalytic chemical reaction with the predicted catalyst or ligand. A network interface can be provided by the computer through which a user uploads the result data of a chemical reaction from the performed experiment. Embodiments of the technology may include hardware, methods or processes, or computer tangible media.

[0039] In one general aspect, the system disclosed herein can select a chemical space for grouping, where each chemical space is defined by a plurality of potential chemicals; select a plurality of chemical features, where each chemical feature corresponds to a plurality of potential chemicals, group the plurality of potential chemicals based on the plurality of chemical features in the grouping space; select a plurality of representative chemicals from the potential chemicals, where each representative chemical corresponds to a group of the plurality of potential chemicals grouped within the grouping space; and / or recommend a test suite with a plurality of test chemicals, where each test chemical corresponds to one of the plurality of representative chemicals. Other embodiments of this aspect include corresponding computer systems, devices, and computer programs recorded on one or more computer storage devices, each configured to perform the operations of the method.

[0040] In one general aspect, the disclosed system can receive a plurality of results from a plurality of experiments performed using a plurality of test chemicals, where each of the plurality of results corresponds to the corresponding result of the corresponding test chemical among the plurality of test chemicals. The system can also recommend the best catalyst or best ligand among all available catalysts or ligands for a catalytic chemical reaction in the prediction space according to the prediction. Other embodiments of this aspect include corresponding computer systems, devices, and computer programs recorded on one or more computer storage devices, each configured to perform the operations of the method.

[0041] In one general aspect, the method may include performing a preliminary screening experiment on a set of test chemicals to generate results. The method may also include obtaining chemical features that describe the properties of potential chemicals. The method may also include reducing the dimensionality of the chemical features to generate a predetermined number of different chemical spaces. Additionally, the method may include calculating the distance between each of the predetermined number of chemical spaces and each test chemical for a given potential chemical. The method may also include normalizing the distances for each of the predetermined number of chemical spaces to the interval [0, 1]. The method may also include subtracting the normalized distances from 1 to generate a distance metric for each potential chemical and each test chemical in each chemical space.

[0042] In one general aspect, the method may include embodiments that may include one or more of the following features as part of the method. The method may also include averaging the distance metrics over all chemical spaces for each potential chemical. The method may also include normalizing the results obtained from the preliminary screening experiment to [0, 1] and converting them to weights. The method may also include multiplying the weights by the distance matrix to obtain a weighted distance matrix, which is generated by the averaging operation and has potential chemical axes and test chemical axes. The method may also include taking the maximum value along the potential chemical axis to obtain a score for each potential chemical. The method may further include sorting the N potential chemicals from highest to lowest based on the scores obtained. Other embodiments of this aspect include corresponding computer systems, devices, and computer programs recorded on one or more computer storage devices, each configured to perform the operations of the method.

[0043] Embodiments may include one or more of the following features. The chemical features may be DFT-based features. PCA, spatial PCA, kernel PCA with an RBF kernel, kernel PCA with a cosine kernel, fast ICA, spectral embedding, isometric mapping, or local linear embedding may be used to reduce the chemical features. The results obtained from the preliminary screening experiment may be yield or enantioselectivity. The weights obtained from the normalized results are multiplied column-wise by the distance matrix to obtain a weighted distance matrix. The chemical features are the properties of phosphine ligands. Embodiments of the techniques may include hardware, methods or processes, or computer tangible media.

[0044] In one general aspect, the method may include: obtaining a list of N potential chemicals; obtaining chemical features configured to describe the properties of the N potential chemicals; clustering the N potential chemicals based on the chemical features of the N potential chemicals to obtain a set of representative chemicals, the set of representative chemicals defining a set of test chemicals; calculating the distance between each test chemical and each representative chemical using a distance metric based on the chemical features; normalizing the distances to the interval [0, 1]; subtracting the normalized distances from 1 to generate a distance metric for each representative chemical and each test chemical; normalizing the results obtained from a preliminary screening experiment to [0, 1] and converting them to weights; multiplying the weights by the distance matrix to obtain a weighted distance matrix; taking the maximum value along the representative chemical axis to obtain a score for each representative chemical; and / or ranking the N potential chemicals based on the scores obtained for their representative chemicals. Other implementations of this aspect include corresponding computer systems, devices, and computer programs recorded on one or more computer storage devices, each configured to perform the operations of the method.

[0045] Implementations may include one or more of the following features. A computer program product may be implemented as the systems or methods described herein (as a computer-readable medium and / or a non-transitory computer-readable medium, etc.). In some implementations, the distances may be calculated after dimensionality reduction of the chemical features. Implementations of the techniques may include hardware, methods or processes, or computer tangible media.

[0046] In one general aspect, the method may include obtaining a list of N potential chemicals. The method may also include receiving the results of M potential chemicals, thereby defining test chemicals; obtaining chemical features configured to describe the properties of the N potential chemicals; determining a score for each representative chemical; and ranking the N potential chemicals based on the scores obtained for their representative chemicals. Other implementations of this aspect include corresponding computer systems, devices, and computer programs recorded on one or more computer storage devices, each configured to perform the operations of the method.

[0047] Embodiments may include one or more of the following features. The act of determining a score for each representative chemical may include: calculating the distance between each test chemical and each representative chemical using a distance metric for each in a multi-dimensional reduced space of chemical features; normalizing the distances to the interval [0, 1] for each of the multi-dimensional reduced spaces; subtracting the normalized distances from 1 to generate a distance metric for each representative chemical and each test chemical for each of the multi-dimensional reduced spaces; averaging the distance metrics for each representative chemical and each test chemical across all the dimensionality-reduced spaces; normalizing the results obtained from the received results of M potential chemicals to [0, 1] and converting it to weights; multiplying the weights by the distance matrix to obtain a weighted distance matrix having representative chemical axes and test chemical axes; and / or taking the maximum along the representative chemical axis to obtain a score for each representative chemical. The test chemicals may be removed from the representative chemical axis. Embodiments of the techniques may include hardware, methods or processes, or computer tangible media.

[0048] Brief Description of the Drawings

[0049] These and other aspects will become more apparent from the following detailed description of various embodiments of the disclosure with reference to the accompanying drawings, in which:

[0050] Figure 1 A schematic overview of a system for optimizing chemical reactions using machine learning in accordance with an embodiment of the disclosure is shown;

[0051] Figure 2 A phosphine ligand clustering in accordance with an embodiment of the disclosure is shown;

[0052] Figure 3 A clustering of phosphine ligands based on molecular characterization and groups of molecular descriptors in accordance with an embodiment of the disclosure is shown;

[0053] Figure 4 The use of a physical kit in accordance with an embodiment of the disclosure is shown, the physical kit containing 24 phosphine ligands from each cluster representing the entire chemical space;

[0054] Figure 5 An AI-based web platform for recommending one or more optimal catalysts for reacting with a specific group of reactants in accordance with an embodiment of the disclosure is shown; and

[0055] Figure 6 A system for selecting one or more optimal chemicals in accordance with an embodiment of the disclosure is shown.

[0056] Detailed Description

[0057] Figure 1 FIG. 1 shows a schematic overview of a system 7 for finding the best catalyst for a desired catalytic chemical reaction using an assembled test set. The catalyst selection system 7 includes a web-based platform 5, implemented, for example, as a homepage, which runs on a computer 3 in the form of a server and which can be accessed by a user 1 via a web browser on a remote device (such as a mobile phone, a personal computer, etc.). Thus, the web-based platform 5 provides a user interface 4, preferably a graphical user interface (GUI), through which the user 1 can input the results of the experiments he has performed by using the assembled test set. In addition, based on the information given by the user 1 (such as the desired catalytic chemical reaction), the web-based platform can also be configured to first provide the function of assembling the test set via the homepage. Using this information and further data on the availability of specific catalysts or ligands, the system 7 can calculate the test set components and submit a request for assembling the test set.

[0058] For example, in one embodiment of the present disclosure, the "Phosphine Predictor" recommends phosphine ligands for cross-coupling reactions (such as Suzuki and Buchwald catalysis). For C-C or C-N cross-coupling reactions of a specific reaction substrate, it is recommended that commercially available monodentate phosphine ligands should be tried. The user runs the reaction using the provided designated general training set ligands (test set) and inputs the results (which can be yield, efficiency, or cost-effectiveness) into a secure Phosphine Predictor portal (such as via the user interface 4). The recommendation of the optimal ligand will be uploaded to the user account.

[0059] The system 7 also includes a machine learning algorithm 6 as described herein. The machine learning algorithm 6 can be designed to learn patterns and make predictions or decisions based on the given data. The machine learning algorithm 6 can use datasets of various chemicals as described herein. The model can be used to make predictions for hypothetical chemical reactions. For new inputs, the machine learning algorithm can make predictions based on the patterns it has previously learned or using any algorithm as described herein. The system 7 can perform parallel computations on the computer 3 using CPUs and / or GPUs.

[0060] This exemplary embodiment will be further introduced by describing method actions, which further explain specific instances of the method of the present disclosure. The method in one embodiment includes being in an AI-based experimental design platform. The embodiments are not limited to the hardware disclosed and used herein.

[0061] Figure 2 FIG. 2 shows an overview of all three steps of the method of the present disclosure: clustering of phosphine ligands based on molecular characterization and a set of molecular descriptors (step 1), a physical set containing 24 phosphine ligands from each cluster (step 2), and an AI-based web platform for recommending the optimal catalyst for a reaction with a specific set of reactants (step 3).

[0062] Figure 3 The clustering of phosphine ligands by k-means clustering method is then shown to screen all ligands available on the corresponding used web platform or homepage. The catalysts / ligands are parameterized with chemical features from cheminformatics and computational modeling covering the chemical space. The catalysts or ligands are then grouped into 24 clusters based on chemical characterization and molecular descriptors.

[0063] exist Figure 4 In the example, one representative catalyst / ligand from each of the 24 clusters may be included in a physical set that will be available for purchase through a web-based platform 5. The selection is based on source availability. Potential customers obtain the set and perform test experiments with all 24 representative catalysts / ligands from the physical set for their specific catalytic chemical reaction of interest, keeping all other reaction conditions constant (if possible).

[0064] The customer enters the yield values ​​of 24 chemical reactions into a web portal or interface of the web-based platform 5, such as Figure 5 Based on this input, the system performs regression to interpolate between 24 clusters in the chemical space of catalysts / ligands and recommends the best possible catalyst / ligand for a given reaction.

[0065] Figure 6 A system 600 is shown for suggesting potential chemicals that may be expected to optimize a chemical reaction based on a combination of distance metrics and performance results. The system 600 may be used in Figure 1 The system 600 is implemented on a computer 3. The system 600 can be implemented in hardware, software, software executed by a processor and / or GPU, etc. In some embodiments, the system 600 can be implemented on the cloud as a software-as-a-service platform, as a distributed system, etc. The system 600 includes an interface component 614 that retrieves chemical features 616 from a database 612 and a network interface component 602 that receives results 604 from a test chemical uploaded by a user.

[0066] The interface component 614 obtains chemical features 616 that describe the properties of a given set of potential chemicals. These chemical features 616 can be obtained using a standard interface (such as a network interface, REST API, etc.). In a particular embodiment, the chemical features used are DFT-based features, as described in Gensch, T.; dos Passos Gomes, G.; Friederich, P.; Peters, E.; Gaudin, T.; Pollice, R.; Jorner, K.; Nigam, A.; Lindner-D'Addario, M.; Sigman, M.S; Aspuru-Guzik, A.; A Comprehensive Discovery Platform for Organophoslus Ligands for Catalysis. J. Am. Chem. Soc. 2022, 144, 3, 1205–1217, the content of which is incorporated herein by reference in its entirety. However, any other type of chemical feature that describes the properties of potential chemicals can also be used. For example, phosphine ligand properties related to optimizing chemical reactions using these chemicals (such as catalysts) can also be considered.

[0067] Due to the high dimensionality of the chemical features 616, dimensionality reduction is applied to the chemical features by the dimensionality reduction component 618. In a particular embodiment of the present disclosure, dimensionality reduction is applied nine times using different techniques to generate nine different chemical spaces 609. For ease of viewing, two chemical spaces 608, 610 are shown as spanning from space p to space q. Any suitable number of chemical spaces 609 can be used. As an alternative embodiment, other dimensionality reduction techniques (such as autoencoders and t-SNE) can also be employed. The purpose of dimensionality reduction is to reduce the number of features to a manageable number, which can be used efficiently and effectively to compare the properties of potential chemicals.

[0068] As mentioned, the chemical features 616 are reduced to nine different embeddings 620, which can be mapped to nine different chemical spaces 609 using the space mapping component 622. Dimensionality reduction can be accomplished using PCA, spatial PCA, kernel PCA with an RBF kernel, kernel PCA with a cosine kernel, fast ICA, spectral embedding, isometric mapping, locally linear embedding, and multidimensional scaling. In certain particular embodiments, each of the embeddings 620 can be 20-dimensional.

[0069] Once the chemical features 616 have been reduced and mapped by the spatial mapping component 622, the distance from each potential chemical within each generated chemical space 609 to a given set of test chemicals is calculated. The distances are calculated using the reduced chemical features, which describe the properties of the potential chemicals relevant to the optimized chemical reaction. If there are 24 test chemicals and 9 spaces 609, then 24 times 9 distances are calculated, resulting in a total of 216 distance terms for each potential chemical.

[0070] Then the distances for each of the nine chemical spaces are normalized to the interval [0, 1] (e.g., Euclidean distance). The normalization can be for each space of the space 609 (e.g., because the absolute value highly depends on that space). For example, in a particular embodiment, there can be 9 sets of MxN distances, where each set of MxN distances can be normalized to 0, 1. Any normalization can be used, such as linear normalization, min-max normalization, etc. This allows for the standardization of the distances and makes the results 604 easier to compare and analyze. Subsequently, all distances are subtracted from 1 to generate a distance metric, where the shortest distance within each chemical space is now 1 and the maximum distance is 0. As an alternative embodiment, other types of distance metrics can also be used (e.g., Mahalanobis distance, Manhattan distance, Euclidean distance, etc.). The choice of distance metric depends on the nature of the chemical properties being compared and the specific application. Additionally, alternatively or optionally, any post-normalization technique can also be employed, such as rescaling or standardizing the distances.

[0071] Next, the average of all the distance metrics is taken for each potential chemical across the different embeddings (i.e., chemical spaces 609). This generates an NxM matrix, where N is the number of potential chemicals for which recommendations are being made, and M is the number of test chemicals (e.g., M = 24). Each position within the matrix may represent the average (mean) of the nine distance metrics for a particular potential chemical and a particular test chemical across all the spaces 609. As an alternative embodiment, other average calculation techniques (e.g., weighted average or geometric average) can be used to perform this step.

[0072] Results 604 from experiments performed using test chemicals can be uploaded by network interface component 602. Results 604 (e.g., yields) are also normalized to [0,1] and converted to weights using an exponential function of the form 2^(x-1), where x is the result. These weights are then multiplied by the distance matrix column by column. The maximum value is taken along the potential chemical axis, resulting in a score ranging from 0 to 1 for each potential chemical, representing the best distance / performance combination relative to the test chemical. These values ​​are considered scores 624. As an alternative embodiment, different score calculation methods can also be used, such as a weighted sum or product of distance and performance scores.

[0073] These scores 624 are then used to rank the N potential chemicals from high to low, where the higher the value, the better the predicted result. The disclosed method provides a reliable and effective method to suggest potential chemicals that are expected to optimize chemical reactions. By taking the potential chemicals with the highest scores, another set of test chemicals can be proposed, and the process can be repeated.

[0074] As a further alternative embodiment, the disclosed method can be used for various industries, including pharmaceuticals, materials science and agricultural chemicals. The method can also be used to predict the activity of other chemical compounds other than potential chemicals, such as drug candidates or natural products. These chemical features can be obtained using various sources, including computer simulations (insilico) or in vitro tests (in vitro assays). In addition, in addition to the nine different dimensionality reduction techniques used in the present disclosure, other techniques, such as UMAP or non-negative matrix factorization, can also be used.

[0075] In addition, an alternative embodiment of the distance metric calculation step is the use of a machine learning model, such as a regression model or a neural network. These models can predict the distance between a potential chemical and a test chemical based on chemical features. The prediction accuracy of these models can be evaluated using cross validation or testing on reserved data.

[0076] Another alternative embodiment of the score calculation step is to incorporate uncertainty measures (such as confidence intervals or probability distributions of the scores). These measures can provide additional information about the reliability and robustness of the score predictions.

[0077] In summary, the disclosed methods and systems provide a general and comprehensive approach for predicting the most promising potential chemicals for optimizing chemical reactions. The method allows the use of various chemical features and dimensionality reduction techniques, as well as alternative distance metrics and score calculation methods. These embodiments make the method applicable to various chemical industries and can improve the accuracy and reliability of predictions.

[0078] The present invention is further illustrated by the following examples hereinafter, which should not be construed as limiting. Those skilled in the art should recognize that various modifications, additions, and changes can be made to the present invention without departing from the spirit and scope of the invention defined in the appended claims. Examples

[0079] General Information

[0080] Unless otherwise stated, all reagents and solvents were purchased from MilliporeSigma and used as received. 2'-dicyclohexylphosphino-2-methoxy-1-phenylnaphthalene, cBRIDP, and di-tert-butyl(2',6'-dimethoxy-[1,1'-biphenyl]-2-yl)phosphine were purchased from Ambeed. VPhos, tris(m-tolyl)phosphine, 9-[2-(dicyclohexylphosphino)phenyl]-9H-carbazole, CPhos, trioctylphosphine, and bis(3,5-bis(trifluoromethyl)phenyl)(2',6'-bis(isopropoxy)-3,6-dimethoxybiphenyl-2-yl)phosphine were purchased from STREM. 3-(Diphenylphosphino)phenol and 2-(dicyclohexylphosphino)-2'-methoxybiphenyl were purchased from Combi-Blocks. 2-Diphenylphosphino-6-methylpyridine and tris(diethylamino)phosphine were purchased from TCI.

[0081] Flash chromatography was performed on a Biotage Isolera TM or Biotage Selekt system. Compounds were characterized by 1 1H NMR and 13 13C NMR. NMR spectra were recorded on a Varian 500 MHz instrument or a Bruker 500 MHz instrument. All 1 1H NMR experiments were reported in δ units (parts per million (ppm)) and measured relative to the signal of residual deuterated chloroform (7.26 ppm), and all 13 13C NMR spectra were reported in ppm relative to deuterated chloroform (77.23 ppm) and all were obtained with 1 1H decoupling. All GC analyses were performed on an Agilent 7820A gas chromatograph with an FID detector using an SPB-1 fused silica column 30 m x 250 μm x 1 μm (cat#: 24029). All reaction flasks were prepared in a positive pressure Vac Omni-Lab glove box and reacted on a Radley's Mya4 reaction station under positive pressure nitrogen. Note! Pure phosphines may be pyrophoric as they may react with air and moisture; however, when used as solutions, the pyrophoricity can be minimized. Follow all precautions in the SDS.

[0082] Phosphine set ligands

[0083]

[0084]

[0085]

[0086]

[0087] Table S1. 24 representative phosphine prediction set ligands and structures.

[0088] Ligand prediction model

[0089] The following reference Figure 6 describes a specific implementation of the above system 600. To understand how the ligand prediction / recommendation model described herein works, especially with reference to Figure 6 , we will use the term "set ligand" (a specific implementation of a test chemical) to denote any ligand in the set of 24 ligands for which a preliminary screening experiment is conducted, and for any ligand that is not part of that set and for which we obtain a ranking via our model to recommend for a new experiment (a specific implementation of a potential chemical), we will use only "ligand".

[0090] The model (e.g., system 600) recommends ligands based on two criteria, namely the distance to the set ligand (how similar the ligand is to the set ligand) and the performance of that set ligand (yield / conversion in our case, but could be enantioselectivity or any other criterion).

[0091] The distances are obtained from molecular features / descriptors that characterize the ligand properties. In our case, we used DFT-based features as described in the kraken paper (Gensch, T.; dos Passos Gomes, G.; Friederich, P.; Peters, E.; Gaudin, T.; Pollice, R.; Jorner, K.; Nigam, A.; Lindner-D'Addario, M.; Sigman, M. S.; Aspuru-Guzik, A.; A Comprehensive Discovery Platform for Organophosphorus Ligands for Catalysis. J. Am. Chem. Soc. 2022, 144, 3, 1205–1217). These features can be obtained from the website (https: / / kraken.cs.toronto.edu / ) using the REST API or can be computed following the code as published in the corresponding GitHub repository (https: / / github.com / aspuru-guzik-group / kraken). In principle, any other set of descriptors that characterize the properties of phosphine ligands relevant to catalysis can be used. Due to the high dimensionality of the features, dimensionality reduction is applied to yield an embedding with 20 components (e.g., embedding 620). Again, various techniques can be used for this - in total, we used nine dimensionality reduction methods (PCA, spatial PCA (SparsePCA), kernel PCA with RBF kernel and with cosine kernel, FastICA, spectral embedding, Isomap, locally linear embedding, and MDS). For each of these embeddings, the distance of each ligand to each set of ligands is computed. For each embedding, these distances are normalized to the [0,1] interval as described above and then the distance is also subtracted from 1 as described above. The resulting distances are such that 1 corresponds to the closest ligand (identity of the set of ligands) and 0 is the farthest ligand in the set of ligands. Finally, the average of all distances from different embeddings is taken. This yields an NxM distance matrix, where N is the number of ligands for which recommendations are made and M is the number of set of ligands (M = 24).

[0092] This performance (here yield / conversion) is also normalized to [0,1] and transformed to weights via an exponential function of the form 2 x-1 where x is the performance, yielding 24 weights for the 24 sets of ligands. These weights are then multiplied column-wise by the distance matrix and then the maximum is taken along the ligand axis, yielding a score between 0 and 1 for each ligand for the best distance / performance combination relative to the set of ligands.

[0093] Then, using the final ranking of the N ligands, new experiments are proposed by taking the ligand with the highest score from the above procedure. In principle, any list of N ligands for which features can be obtained or calculated features can be used. In our case, a list of approximately 400 monodentate phosphine ligands was used, and we also used this list for initial clustering to construct ligand sets.

[0094] Cross-coupling screening reaction

[0095] Example 1

[0096] General procedure for Buchwald-Hartwig C-N cross-coupling reaction 1:

[0097]

[0098] N-[2,6-bis(1-methylethyl)phenyl]-2,4,6-tris(1-methylethyl)aniline (1): A 4 mL screw-cap vial was placed in a glove box and charged with Pd2(dba)3 (4.6 mg, 0.5 mmol, 0.5 mol %), phosphine ligand (0.1 mmol, 1.0 mol %), 4,4'-di-tert-butylbiphenyl (internal standard, 80 mg, 0.3 mmol, 0.3 equiv), NaO-t-Bu (144 mg, 1.5 mmol, 1.5 equiv), 2,4,6-triisopropylbromobenzene (0.25 mL, 1.0 mmol, 1.0 equiv), 2,6-diisopropylaniline (0.23 mL, 1.2 mmol, 1.2 equiv), and toluene (2.0 mL, 0.5 M). The vial was sealed with a rubber / teflon septum and removed from the glove box, and then the reaction was placed into a Radley Mya4 reaction station preheated to 80 °C with the stirring set at 300 rpm. After a reaction time of 1 h or 20 h, 50 μL of the sample was diluted with 1 mL of EtOAc via a syringe filter and the reaction mixture was analyzed by GC. After the reaction was judged to be complete by GC, the reaction mixture was diluted with ethyl acetate and filtered through a short plug of silica gel. After drying, the crude reaction mixture was dry loaded onto silica gel (0-20% EtOAc / heptane) for Biotage flash chromatography to afford N-[2,6-bis(1-methylethyl)phenyl]-2,4,6-tris(1-methylethyl)aniline as a colorless solid. The product was confirmed by comparison with literature NMR spectral data (Raders, S.M.; Moore, J.N.; Parks, J.K.; Miller, A.D.; Leiβing, T.M.; Kelley, S.P.; Rogers, R.D.; Shaughnessy, K.H. Trineopentylphosphine: A Conformationally Flexible Ligand for the Coupling of Sterically Demanding Substrates in the Buchwald–Hartwig Amination and Suzuki–Miyaura Reaction. J. Org. Chem. 2013, 78, 4649-4664).

[0099] 11H NMR (CDCl3): δ 7.7 (d, J = 7.6 Hz, 2H), 6.98 - 6.93 (m, 3H), 4.77 (s, 1H), 3.16 - 3.2 (m, 4H), 3.7 (sept, J = 6.7 Hz, 1H), 1.24 (d, J = 6.9 Hz, 6H), 1.8 (t, J = 7.1 Hz, 24H).

[0100] 13 13C NMR (CDCl3): δ 143.8, 141.9, 141.2, 140.1, 138.3, 124.0, 122.3, 121.8, 34.2, 28.1, 27.9, 24.5, 23.9, 23.8.

[0101] Buchwald - Hartwig C - N cross - coupling reaction 1 a of ligand screening set

[0102]

[0103]

[0104] Table 2. Ligand screening set for Buchwald - Hartwig C - N cross - coupling reaction 1 a of ligand screening. a Conditions: Ar - Br (1.0 mmol), Ar - NH2 (1.2 mmol), Pd2dba3 (0.5 mol%), ligand (1.0 mol%), NaO t Bu (1.5 mmol), toluene (0.5 mol), 80 °C, 1 or 20 h. b Average of 2 runs, % conversion determined by GC.

[0105] Predicted ligand structure for Buchwald - Hartwig C - N cross - coupling reaction 1

[0106]

[0107]

[0108] Table S3. Predicted ligand structure for Buchwald - Hartwig C - N cross - coupling reaction 1.

[0109] Buchwald - Hartwig C - N cross - coupling reaction 1 a of predicted ligand screening results

[0110]

[0111] Table S4 Predicted ligand screening results for Buchwald-Hartwig C-N cross-coupling reaction 1 a a Conditions: Ar-Br (1.0 mmol), Ar-NH2 (1.2 mmol), Pd2dba3 (0.5 mol%), ligand (1.0 mol%), NaO t Bu (1.5 mmol), toluene (0.5 mol), 80 °C, 1 or 20 h. b GC conversion rate.

[0112] Example 2

[0113] General procedure for Buchwald-Hartwig C-N cross-coupling reaction 2

[0114]

[0115] ​2-(1H-Indol-1-yl)benzoxazole (2): A 4 mL screw-cap vial was placed in a glove box and charged with Pd2(dba)3 (4.6 mg, 0.5 mmol, 0.5 mol%), phosphine ligand (0.1 mmol, 1.0 mol%), 4,4'-di-tert-butylbiphenyl (internal standard, 80 mg, 0.3 mmol, 0.3 equiv), NaO-t-Bu (144 mg, 1.5 mmol, 1.5 equiv), indole (141 mg, 1.2 mmol, 1.2 equiv), K3PO4 (318 mg, 1.5 mmol, 1.5 equiv), toluene (2.0 mL, 0.5 M), and 2-chlorobenzoxazole (114 μL, 1.0 mmol, 1.0 equiv). The vial was sealed with a rubber / teflon septum and removed from the glove box, and then the reaction was placed in a Radley Mya4 reaction station preheated to 80 °C with stirring set at 300 rpm. After 16 h, 50 μL of the sample was diluted with 1 mL of EtOAc through a syringe filter and the reaction mixture was analyzed by GC. As judged by GC, after the reaction had reached completion, the reaction mixture was diluted with ethyl acetate and filtered through a short column of silica gel. After drying, the crude reaction mixture was loaded onto silica gel for Biotage flash chromatography (0–5% EtOAc / heptane) to afford 2-(1H-indol-1-yl)benzoxazole as a colorless solid. The product was confirmed by comparison with the literature NMR spectral data (Li, D.-H.; Lan, X.-B.; Song, A.-X.; Rahman, M. M.; Xu, C.; Huang, F.-D.; Szostak, R.; Szostak, M.; Liu, F.-S. Buchwald-Hartwig Amination of Coordinating Heterocycles Enabled by Large-but-Flexible Pd-BIAN-NHC Catalysts. Chem. Eur. J. 2022, 28, e202103341).

[0116] 1 1H NMR (CDCl3) δ 8.57 (d, J = 8.3 Hz, 1H), 7.88 (d, J = 3.6 Hz, 1H), 7.73–7.64 (m, 2H), 7.55 (d, J = 7.8 Hz, 1H), 7.49–7.41 (m, 1H), 7.33 (m, 3H), 6.79 (d, J = 3.6 Hz, 1H).

[0117] 1313C NMR (CDCl3) δ 154.8, 148.5, 141.5, 134.7, 130.2, 124.9, 124.7, 124.6, 123.6, 123.0, 121.3, 118.9, 114.6, 109.9, 108.5.

[0118] Buchwald-Hartwig C-N Cross-Coupling Reaction 2 a Ligand Screening for a Set of Ligands

[0119]

[0120] Ligand <![CDATA[% Conversion Rate b > Ligand <![CDATA[% Conversion Rate b > L1 6.17 L13 18.46 L2 5.16 L14 12.47 L3 18.61 L15 7.22 L4 7.48 L16 3.45 L5 4.50 L17 25.91 L6 6.91 L18 10.38 L7 6.40 L19 4.82 L8 6.44 L20 5.79 L9 4.61 L21 9.56 L10 5.99 L22 11.45 L11 4.86 L23 5.74 L12 22.56 L24 7.80

[0121] Table S5. Buchwald-Hartwig C-N Cross-Coupling Reaction 2 a Ligand Screening for a Set of Ligands. a Conditions: Ar-Cl (1.0 mmol), indole (1.2 mmol), Pd2dba3 (0.5 mol%), ligand (1.0 mol%), K3PO4 (1.5 mmol), toluene (0.5 mol), 80 °C, 16 h. b Average of 2 runs, % conversion determined by GC.

[0122] Predicted Ligand Structures for Buchwald-Hartwig C-N Cross-Coupling Reaction 2

[0123]

[0124]

[0125] Table S6. Predicted Ligand Structures for Buchwald-Hartwig C-N Cross-Coupling Reaction 2.

[0126] Buchwald-Hartwig C-N Cross-Coupling Reaction 2 a Predicted Ligand Screening Results

[0127] Ligand <![CDATA[% conversion rate b > PL11 12.63 PL12 11.51 PL13 16.43 PL14 21.40 PL15 15.24 PL16 13.67 PL17 10.37 PL18 5.26 PL19 12.43 PL20 11.37

[0128] Table S7 Buchwald-Hartwig C-N Cross-Coupling Reaction 2 a Predicted Ligand Screening Results. a Conditions: Ar-Cl (1.0 mmol), indole (1.2 mmol), Pd2dba3 (0.5 mol%), ligand (1.0 mol%), K3PO4 (1.5 mmol), toluene (0.5 mol), 80 °C, 16 h. bGC conversion rate.

[0129] Using RuPhos a Optimizing reaction 2

[0130]

[0131]

[0132]

[0133] Table S8. Using RuPhos a Optimization of reaction 2. a Standard conditions: Ar-Cl (1.0 mmol), indole (1.2 mmol), Pd2dba3 (0.5 mol%), RuPhos (1.0 mol%), base (1.5 mmol), solvent (0.5 mol), 16 h. b GC conversion rate. c Indole (4 equiv). d ACN (1 equiv).

[0134] Example 3

[0135] General procedure for Suzuki C-C cross-coupling reaction 3:

[0136]

[0137] 2-(2-Thienyl)quinoxaline (3): A 4 mL screw-cap vial was placed in a glove box and charged with Pd2(dba)3 (4.6 mg, 0.5 mmol, 0.5 mol %), phosphine ligand (0.1 mmol, 1.0 mol %), 4,4'-di-tert-butylbiphenyl (internal standard, 80 mg, 0.3 mmol, 0.3 equiv), 2-thienylboronic acid (192 mg, 1.5 mmol, 1.5 equiv), 2-chloroquinoxaline (165 mg, 1.0 mmol, 1.0 equiv), toluene (2.0 mL, 0.5 M), and Et3N (420 μL, 3.0 mmol, 3.0 equiv). The vial was sealed with a rubber / teflon septum and removed from the glove box, and the reaction was then placed in a Radley Mya4 reaction station preheated to 100 °C with stirring set to 300 rpm. After 20 h, 50 μL of the sample was diluted with 1 mL of EtOAc through a syringe filter, and the reaction mixture was analyzed by GC. As judged by GC, after the reaction had reached completion, the reaction mixture was diluted with ethyl acetate and filtered through a short silica column. After drying, the crude reaction mixture was dry loaded onto silica gel for Biotage flash chromatography (0 - 30% EtOAc / heptane) to afford 2-(2-thienyl)quinoxaline as a white solid. The product was confirmed by comparison with literature NMR spectral data. (Knapp, D.M.; Gillis, E.P.; Burke, M.D. A General Solution for Unstable Boronic Acids: Slow-Release Cross-Coupling from Air-Stable MIDA Boronates. J. Am. Chem. Soc. 2009, 131, 20, 6961–6963)

[0138] 1 1H NMR (500.1 MHz, CDCl3): δ 9.25 (s, 1H), 8.9 - 8.7 (m, 2H), 7.88 - 7.87 (m, 1H), 7.78 - 7.69 (m, 2H), 7.56 - 7.55 (m, 1H), 7.23 - 7.21 (m, 1H).

[0139] 13 13C NMR (CDCl3): δ 147.3, 142.2, 142.1, 142.0, 141.3, 130.4, 129.7, 129.1, 129.1, 128.4, 126.

[0140] Kit ligand screening for Suzuki C-C cross-coupling reaction 3 a of

[0141]

[0142] Ligand <![CDATA[% Conversion Rate b > Ligand <![CDATA[% Conversion Rate b > L1 0.00 L13 31.70 L2 5.80 L14 1.85 L3 4.4 L15 15.21 L4 26.33 L16 26.9 L5 21.77 L17 52.42 L6 3.76 L18 3.56 L7 0.00 L19 3.51 L8 28.76 L20 3.68 L9 2.48 L21 45.21 L10 2.35 L22 27.87 L11 16.00 L23 3.37 L12 27.19 L24 9.68

[0143] Table S9 for ligand screening in Suzuki C-C cross-coupling reaction 3 a of the ligand set. a Conditions: Ar-Cl (1.0 mmol), Ar-B(OH)2 (1.5 mmol), Pd2dba3 (0.5 mol%), ligand (1.0 mol%), Et3N (3.0 mmol), toluene (0.5 mol), 100 °C, 20 h. b Average of 2 runs, % conversion determined by GC.

[0144] Predicted ligand structures for Suzuki C-C cross-coupling reaction 3

[0145]

[0146]

[0147] Table S10. Predicted ligand structures for Suzuki C-C cross-coupling reaction 3.

[0148] For Suzuki C-C cross-coupling reaction 3 a of the predicted ligand screening results

[0149] Ligand <![CDATA[% Conversion Rate b > PL21 49.96 PL22 2.4 PL23 24.67 PL24 37.82 PL25 49.75 PL26 43.36 PL27 14.69 PL28 45.67 PL29 42.97 PL30 11.63 PL31 0 PL32 31.61 PL33 16.9 PL34 20.39

[0150] Table S11. For Suzuki C-C cross-coupling reaction 3 a of the predicted ligand screening results. a Conditions: Ar-Cl (1.0 mmol), Ar-B(OH)2 (1.5 mmol), Pd2dba3 (0.5 mol%), ligand (1.0 mol%), Et3N (3.0 mmol) ， toluene (0.5 mol), 100 °C, 20 h. b GC conversion.

[0151] Using RuPhos a to optimize Suzuki C-C cross-coupling reaction 3

[0152]

[0153] Entry Solvent Base (equivalent) <![CDATA[% Conversion Rate b > STD Toluene <![CDATA[Et3N(3.0)]]> 49.75 1 Toluene <![CDATA[K3PO4(3.0)]]> 33.89 2 Toluene <![CDATA[Cs2CO3(3.0)]]> 4.47 3 1,4 - Dioxane <![CDATA[Et3N(3.0)]]> 28.17 4 <![CDATA t BuOH]]> <![CDATA[Et3N(3.0)]]> 55.36 <![CDATA[5 c > 10:1 Toluene:Water <![CDATA[Et3N(3.0)]]> 55.11 <![CDATA[6 d > <![CDATA t BuOH]]> <![CDATA[Et3N(3.0)]]> 50.89

[0154] Table S12. Using RuPhos a to optimize Suzuki C-C cross-coupling reaction 3. aStandard conditions: Ar-Cl (1.0 mmol), Ar-B(OH)2 (1.5 mmol), Pd2dba3 (0.5 mol%), RuPhos (1.0 mol%), base (3.0 mmol), solvent (0.5 mol), 16 h. b GC conversion. c Reaction temperature = 80 °C. d Ar-B(OH)2 (4 equivalents).

[0155] Each of the above features and examples and their combinations can be said to be covered by the present disclosure.

[0156] Accordingly, the present disclosure relates to the following non-limiting aspects:

[0157] (1) A method for optimizing a chemical reaction using machine learning, the method comprising: selecting a chemical space for grouping, each chemical space being defined by a plurality of potential chemicals; selecting a plurality of chemical features, each chemical feature corresponding to a plurality of potential chemicals, grouping the plurality of potential chemicals in the grouping space based on the plurality of chemical features; selecting a plurality of representative chemicals from the potential chemicals, each representative chemical corresponding to a group of the plurality of potential chemicals grouped within the grouping space; and assembling a test set having a plurality of test chemicals, each test chemical corresponding to one of the plurality of representative chemicals.

[0158] (2) The method according to aspect 1, wherein the grouping space is defined by a plurality of chemical features.

[0159] (3) The method according to aspect 1, wherein the grouping space is a dimensionality-reduced space of the plurality of chemical features.

[0160] (4) The method according to aspect 1, the method further comprising: generating a plurality of reduced chemical spaces, wherein each reduced chemical space is a dimensionality-reduced space of the chemical space.

[0161] (5) The method according to aspect 4, the method further comprising: calculating a plurality of distances, wherein each distance of the plurality of distances is a distance between one of the plurality of potential chemicals and one of the plurality of test chemicals.

[0162] (6) The method according to aspect 5, wherein the plurality of test chemicals is a subset of the plurality of potential chemicals.

[0163] (7) The method according to aspect 5, wherein the plurality of potential chemicals does not include the plurality of test chemicals.

[0164] (8) The method according to aspect 4, the method further comprising: calculating a plurality of distance metrics, wherein each distance metric of the plurality of distance metrics is one minus the distance between one of the plurality of potential chemicals and one of the plurality of test chemicals.

[0165] (9) The method according to aspect 8, the method further comprising averaging each of the plurality of distance metrics in all reduced chemical spaces to generate a plurality of average distance metrics.

[0166] (10) The method according to aspect 1, the method further comprising generating a plurality of weights, wherein each weight of the plurality of weights corresponds to one of a plurality of results, and wherein each result of the plurality of results corresponds to one of the plurality of test chemicals.

[0167] (11) The method according to aspect 10, wherein the plurality of weights are determined according to an exponential function.

[0168] (12) The method according to aspect 10, wherein each of the plurality of distances of the distance metric is multiplied by the corresponding weight of the plurality of weights.

[0169] (13) The method according to aspect 12, wherein, among all the multiplied distance metrics between the corresponding potential chemical and the plurality of test chemicals, the maximum value of each corresponding potential chemical is taken.

[0170] (14) The method according to aspect 13, the method further comprising sorting the maximum values of each corresponding potential chemical.

[0171] (15) The method according to aspect 1, the method further comprising testing the plurality of test chemicals to determine a plurality of results, wherein each of the plurality of results corresponds to the corresponding result of one of the plurality of test chemicals.

[0172] (16) The method according to aspect 15, wherein the plurality of results are uploaded to a server.

[0173] (17) The method according to aspect 15, the method further comprising using a plurality of prediction features to select a prediction space to provide a prediction to a result space, wherein the plurality of results define data points within the prediction space mapped to the result space.

[0174] (18) The method according to aspect 17, wherein the result space includes chemical yield.

[0175] (19) The method according to aspect 17, wherein the result space is a by-product metric.

[0176] (20) The method according to aspect 17, wherein the result space is a single parameter.

[0177] (21) The method according to aspect 17, wherein the prediction space includes the grouping space.

[0178] (22) The method according to aspect 17, wherein the prediction space includes the plurality of chemical features therein.

[0179] (23) The method according to aspect 17, wherein the computer is configured to use regression fitting to fit a plurality of results to map the plurality of prediction features to the result space.

[0180] (24) The method according to aspect 21, the method further comprising selecting a chemical from the plurality of potential chemicals corresponding to the optimized value of the result space.

[0181] (25) The method according to aspect 17, wherein the computer is configured to use the plurality of results to train a learner.

[0182] (26) The method according to aspect 25, wherein the learner is at least one of an artificial neural network, K-nearest neighbor, decision tree, random forest, support vector machine, Bayesian regressor, and an ensemble.

[0183] (27) The method according to aspect 1, wherein the plurality of chemical features include a plurality of chemical characterizations.

[0184] (28) The method according to aspect 1, wherein the plurality of chemical features include a plurality of molecular descriptors.

[0185] (29) The method according to aspect 1, wherein the chemical space includes a plurality of catalysts.

[0186] (30) The method according to aspect 1, wherein the chemical space includes a plurality of ligands.

[0187] (31) The method according to aspect 1, wherein the plurality of chemical features are determined by at least one of chemoinformatics and computational modeling on a computer.

[0188] (32) The method according to aspect 1, wherein the plurality of chemical features are stored in a database executed by a computer.

[0189] (33) The method according to aspect 1, the method further comprising: testing the plurality of test chemicals to determine a plurality of results, wherein each of the plurality of results corresponds to a respective result of one of the plurality of test chemicals; using a plurality of predictive features to select a prediction space to provide a prediction to a result space, wherein the plurality of results define data points within the prediction space; using regression to map the plurality of predictive features to the result space; determining, based on the prediction, the best catalyst or ligand among all available catalysts or ligands for catalyzing a chemical reaction in the prediction space; and catalyzing the chemical reaction with the best catalyst or ligand according to the prediction.

[0190] (34) The method according to aspect 1, wherein the grouped space is generated by dimensionality reduction of the plurality of chemical features.

[0191] (35) The method according to aspect 34, wherein the dimensionality reduction is implemented on a computer using principal component analysis.

[0192] (36) The method according to aspect 1, wherein the chemical space is defined by a plurality of phosphine ligands for cross-coupling reactions.

[0193] (37) The method according to aspect 36, wherein the cross-coupling reaction includes one of Suzuki catalysis and Buchwald catalysis.

[0194] (38) The method according to aspect 1, wherein the plurality of test chemicals consists of 24 chemicals.

[0195] (39) The method according to aspect 1, wherein the plurality of chemical features includes at least one of space-filling features, volume features, orientation features, electrical features, input voltage (Vin), frontier molecular orbitals, Fukui functions, NBO analysis, NMR tensors, steric properties, Sterimol L, B1, B5, B1, B5, quadrant analysis, octant analysis, total volume, buried volume, dipole moment, solvation energy, and potential dispersion.

[0196] (40) The method according to aspect 17, wherein the plurality of predictive features includes at least one of space-filling features, volume features, orientation features, electrical features, input voltage (Vin), frontier molecular orbitals, Fukui functions, NBO analysis, NMR tensors, steric properties, Sterimol L, B1, B5, B1, B5, quadrant analysis, octant analysis, total volume, buried volume, dipole moment, solvation energy, and potential dispersion.

[0197] (41)A method for using an experimental test suite, the method comprising: assembling a test suite having a plurality of test chemicals, each test chemical corresponding to one representative chemical among a plurality of representative chemicals; testing the plurality of test chemicals to determine a plurality of results, wherein each of the plurality of results corresponds to a corresponding result of a corresponding test chemical among the plurality of test chemicals; and determining, based on a prediction, the best catalyst or the best ligand among all available catalysts or ligands for a catalytic chemical reaction in a prediction space.

[0198] (42)The method according to aspect 41, the method further comprising: generating a plurality of reduced chemical spaces, wherein each reduced chemical space is a dimension-reduced space of a chemical space.

[0199] (43)The method according to aspect 42, the method further comprising: calculating a plurality of distances, wherein each of the plurality of distances is a distance between a potential chemical among a plurality of potential chemicals and a test chemical among the plurality of test chemicals.

[0200] (44)The method according to aspect 43, further comprising: calculating a plurality of distance metrics, wherein each of the plurality of distance metrics is one minus the distance between a potential chemical among the plurality of potential chemicals and a test chemical among the plurality of test chemicals.

[0201] (45)The method according to aspect 44, further comprising averaging each of the plurality of distance metrics in all of the reduced chemical spaces to generate a plurality of average distance metrics.

[0202] (46)The method according to aspect 41, further comprising generating a plurality of weights, wherein each of the plurality of weights corresponds to one of the plurality of results, and wherein each of the plurality of results corresponds to one of the plurality of test chemicals.

[0203] (47)The method according to aspect 46, wherein the plurality of weights are determined according to an exponential function.

[0204] (48)The method according to aspect 46, wherein each of the plurality of distances of the distance metrics is multiplied by a corresponding one of the plurality of weights.

[0205] (49)The method according to aspect 48, wherein, among all the multiplied plurality of distance metrics between a corresponding potential chemical and the plurality of test chemicals, the maximum value of each corresponding potential chemical is taken.

[0206] (50)The method according to aspect 49, further comprising sorting the maximum values of each corresponding potential chemical to find the best catalyst or the best ligand.

[0207] (51) The method according to aspect 41 further includes using a plurality of predictive features to select a prediction space to provide a prediction to a result space, wherein the plurality of results define data points within the prediction space.

[0208] (52) The method according to aspect 51 further includes using regression to fit data points within the plurality of predictive features to the result space.

[0209] (53) The method according to aspect 51, wherein the plurality of predictive features includes at least one of a space-filling feature, a bulk feature, an orientation feature, an electrical property feature, an input voltage (Vin), a frontier molecular orbital, a Fukui function, NBO analysis, an NMR tensor, a steric property, Sterimol L, B1, B5, B1, B5, quadrant analysis, octant analysis, total volume, buried volume, dipole moment, solvation energy, and potential dispersity.

[0210] (54) The method according to aspect 41 further includes catalyzing a chemical reaction with the best catalyst or the best ligand according to the prediction.

[0211] (55) The method according to aspect 33, wherein the plurality of results are uploaded to a server.

[0212] (56) The method according to aspect 33 further includes using a plurality of predictive features to select a prediction space to provide a prediction to a result space, wherein the plurality of results define data points within the prediction space.

[0213] (57) The method according to aspect 56, wherein the computer is configured to map the plurality of predictive features to the result space using regression by fitting the plurality of results to the predictive features and the result space.

[0214] (58) A method of assembling, by a computer, a test set for optimizing a catalytic chemical reaction, the method including: parameterizing, by a computer, a catalyst or a ligand for a corresponding catalytic chemical reaction for corresponding chemical features specific to the catalytic chemical reaction; grouping the parameterized catalysts or ligands into a given number of clusters based on their chemical features, the clusters covering the chemical space of the catalysts or ligands; using the computer to select a representative catalyst or ligand from each cluster according to a predetermined criterion; and assembling the test set with the selected representative catalysts or ligands as components.

[0215] (59) The method according to aspect 58, wherein the corresponding chemical features are determined by cheminformatics and computational modeling on a computer.

[0216] (60) The method according to aspect 58, further comprising performing a grouping action using a computer by leveraging k-means clustering.

[0217] (61) The method according to aspect 58, wherein the selection is based on a combination of the chemical characteristics of the catalyst or ligand and commercial viability and / or raw material availability as a predetermined criterion.

[0218] (62) The method according to aspect 61, wherein all available catalysts or ligands are stored in a database connected to a computer.

[0219] (63) A test set having a specific given number of catalyst or ligand components and assembled by using the method according to aspect 58.

[0220] (64) The method according to aspect 58, the method further comprising: performing a standard experiment on a catalytic chemical reaction with the components in the test set; inputting the result data of the performed experiment into a computer; using a machine learning regression model running on the computer to interpolate between a given number of clusters in the covered chemical space of all available catalysts or ligands; predicting, in the covered chemical space of all available catalysts or ligands, a catalyst for the best fit for the catalytic chemical reaction using the machine learning regression model; and performing the catalytic chemical reaction using the predicted catalyst or ligand.

[0221] (65) The method according to aspect 64, wherein a network interface is provided by the computer, through which a user uploads the result data of the chemical reaction from the performed experiment.

[0222] (66) A system for optimizing a chemical reaction, the system implemented by an operation group of processor-executable instructions configured to be executed on at least one processor, the at least one processor and the operation group of processor-executable instructions configured to: select a chemical space for grouping, each chemical space defined by a plurality of potential chemicals; select a plurality of chemical characteristics, each chemical characteristic corresponding to the plurality of potential chemicals, and group the plurality of potential chemicals in a grouping space based on the plurality of chemical characteristics; select a plurality of representative chemicals from the potential chemicals, each representative chemical corresponding to a group of the plurality of potential chemicals grouped within the grouping space; and recommend a test set having a plurality of test chemicals, each test chemical corresponding to one of the plurality of representative chemicals.

[0223] (67) A system implemented by a set of processor-executable instructions configured to execute on at least one processor, the at least one processor and the set of processor-executable instructions being configured to: recommend a test suite having a plurality of test chemicals, each test chemical corresponding to one representative chemical among a plurality of representative chemicals; receive a plurality of results from a plurality of experiments using the plurality of test chemicals, wherein each of the plurality of results corresponds to a respective result of a respective test chemical among the plurality of test chemicals; and recommend, based on a prediction, the best catalyst or the best ligand among all available catalysts or ligands for a catalytic chemical reaction in a prediction space.

[0224] (68) A method for suggesting potential chemicals that can be used to optimize a chemical reaction, comprising: performing a preliminary screening experiment on a set of test chemicals to generate results; obtaining chemical features that describe the properties of the potential chemicals; reducing the dimensionality of the chemical features to generate a predetermined number of different chemical spaces; calculating the distance between each of the predetermined number of chemical spaces of a given potential chemical and each of the test chemicals; normalizing the distance in each of the predetermined number of chemical spaces to the interval [0, 1]; subtracting the normalized distance from 1 to generate a distance metric for each potential chemical and each test chemical in each chemical space; averaging the distance metrics for each potential chemical over all chemical spaces; normalizing the results obtained from the preliminary screening experiment to [0, 1] and converting them into weights; multiplying the weights by the distance matrix to obtain a weighted distance matrix, which is generated by the averaging operation and has a potential chemical axis and a test chemical axis; taking the maximum value along the potential chemical axis to obtain a score for each potential chemical; and sorting N potential chemicals from highest to lowest based on the obtained scores.

[0225] (69) The method according to aspect 68, wherein the chemical features are DFT-based features.

[0226] (70) The method according to aspect 68, wherein principal component analysis (PCA), spatial PCA, kernel PCA with an RBF kernel, kernel PCA with a cosine kernel, fast independent component analysis (ICA), spectral embedding, isometric mapping, or local linear embedding is used to reduce the chemical features.

[0227] (71) The method according to aspect 68, wherein the results obtained from the preliminary screening experiment are yield or enantioselectivity.

[0228] (72) The method according to aspect 68, wherein the weights obtained from the normalized results are multiplied by the distance matrix column-wise to obtain a weighted distance matrix.

[0229] (73) The method according to aspect 68, wherein the chemical features are the properties of phosphine ligands.

[0230] (74) A method for suggesting potential chemicals for optimizing a chemical reaction, comprising: obtaining a list of N potential chemicals; obtaining chemical features configured to describe the properties of the N potential chemicals; clustering the N potential chemicals based on their chemical features to obtain a set of representative chemicals, the set of representative chemicals defining a set of test chemicals; calculating the distance between each test chemical and each representative chemical using a distance metric based on chemical features; normalizing the distances to the interval [0, 1]; subtracting the normalized distances from 1 to generate a distance metric for each representative chemical and each test chemical; normalizing the results obtained from a preliminary screening experiment to [0, 1] and converting them to weights; multiplying the weights by the distance matrix to obtain a weighted distance matrix; taking the maximum value along the representative chemical axis to obtain a score for each representative chemical; and ranking the N potential chemicals based on the scores obtained for their representative chemicals.

[0231] (75) A computer program product comprising a non-transitory computer-readable storage medium encoded with instructions for performing the steps of the method according to aspect 74.

[0232] (76) A system for suggesting potential chemicals for optimizing a chemical reaction, comprising a computer system configured to implement the method according to aspect 74.

[0233] (77) The method according to aspect 74, wherein the distance is calculated after dimensionality reduction of the chemical features.

[0234] (78) A computer program product comprising a non-transitory computer-readable storage medium encoded with instructions for implementing the method according to aspect 74.

[0235] (79) A system for identifying potential chemicals for optimizing a chemical reaction, comprising a computer system configured to implement the method according to aspect 74.

[0236] (80) A method for suggesting potential chemicals for optimizing a chemical reaction, comprising: obtaining a list of N potential chemicals; receiving the results of M potential chemicals, thereby defining test chemicals; obtaining chemical features configured to describe the properties of the N potential chemicals; determining a score for each representative chemical; and ranking the N potential chemicals based on the scores obtained for their representative chemicals.

[0237] (81) The method according to aspect 80, wherein the operation of determining the score for each representative chemical includes: calculating the distance between each test chemical and each representative chemical using a distance metric in a space reduced for each of the plurality of dimensions of the chemical characteristics; for each of the spaces reduced for the plurality of dimensions, normalizing the distance to the interval [0, 1]; subtracting the normalized distance from 1 to generate a distance metric for each representative chemical and each test chemical in each of the spaces reduced for the plurality of dimensions; averaging the distance metrics for each representative chemical and each test chemical over all the spaces reduced for the dimensions; normalizing the results obtained from the results of the M received potential chemicals to [0, 1] and converting them into weights; multiplying the weights by the distance matrix to obtain a weighted distance matrix having a representative chemical axis and a test chemical axis; and taking the maximum value along the representative chemical axis to obtain the score for each representative chemical.

[0238] (82) The method according to aspect 81, wherein the test chemical is removed from the representative chemical axis.

[0239] (83) A computer program product comprising a non - transitory computer - readable storage medium encoded with instructions for performing the steps of the method according to aspect 80.

[0240] (84) A system for suggesting potential chemicals that can be used to optimize a chemical reaction, comprising a computer system configured to implement the method according to aspect 80.

Claims

1. A method for optimizing chemical reactions using machine learning, the method comprising: Selecting a chemical space for grouping, each chemical space being defined by a plurality of potential chemicals; Selecting a plurality of chemical features, each chemical feature corresponding to the plurality of potential chemicals, Grouping the plurality of potential chemicals in a grouping space based on the plurality of chemical features; Selecting a plurality of representative chemicals from the potential chemicals, each representative chemical corresponding to a group of the plurality of potential chemicals grouped within the grouping space; And Assembling a test set having a plurality of test chemicals, each test chemical corresponding to one of the plurality of representative chemicals.

2. The method according to claim 1, wherein the grouping space is defined by the plurality of chemical features.

3. The method according to claim 1, wherein the grouping space is a dimensionality-reduced space of the plurality of chemical features.

4. The method according to claim 1, the method further comprising: Generating a plurality of reduced chemical spaces, wherein each reduced chemical space is a dimensionality-reduced space of a chemical space.

5. The method according to claim 4, the method further comprising: Calculating a plurality of distances, wherein each distance in the plurality of distances is a distance between one of the plurality of potential chemicals and one of the plurality of test chemicals.

6. The method according to claim 5, wherein the plurality of test chemicals is a subset of the plurality of potential chemicals.

7. The method according to claim 5, wherein the plurality of potential chemicals does not include the plurality of test chemicals.

8. The method according to claim 4, the method further comprising: Calculating a plurality of distance metrics, wherein each distance metric in the plurality of distance metrics is one minus the distance between one of the plurality of potential chemicals and one of the plurality of test chemicals.

9. The method according to claim 8, the method further comprising averaging each of the plurality of distance metrics in all of the reduced chemical spaces to generate a plurality of average distance metrics.

10. The method according to claim 1, the method further comprising generating a plurality of weights, wherein each weight in the plurality of weights corresponds to one of a plurality of results, wherein each result in the plurality of results corresponds to one of the plurality of test chemicals.

11. The method according to claim 10, wherein the plurality of weights is determined according to an exponential function.

12. The method according to claim 10, wherein each of the plurality of distances of the distance metric is multiplied by a corresponding weight in the plurality of weights.

13. The method according to claim 12, wherein, among all the multiplied distance metrics between the corresponding potential chemical and the plurality of test chemicals, the maximum value of each corresponding potential chemical is taken.

14. The method according to claim 13, the method further comprising sorting the maximum values of each corresponding potential chemical.

15. The method according to claim 1, the method further comprising testing the plurality of test chemicals to determine a plurality of results, wherein each of the plurality of results corresponds to a respective result of one of the plurality of test chemicals.

16. The method according to claim 15, wherein the plurality of results are uploaded to a server.

17. The method according to claim 15, the method further comprising using a plurality of predictive features to select a prediction space to provide a prediction to a result space, wherein the plurality of results define data points within the prediction space that map to the result space.

18. The method according to claim 17, wherein the result space includes chemical yield.

19. The method according to claim 17, wherein the result space is a byproduct metric.

20. The method according to claim 17, wherein the result space is a single parameter.

21. The method according to claim 17, wherein the prediction space includes the grouping space.

22. The method according to claim 17, wherein the prediction space includes the plurality of chemical features therein.

23. The method according to claim 17, wherein a computer is configured to use regression fitting to fit the plurality of results to map the plurality of predictive features to the result space.

Images