Chemical functional homologue search using inconsistent canonicalized query

Abstract

A system and method for providing a chemical similarity search of functional homologs using inconsistently standardized molecular SMILES model inputs or other molecular model inputs to a chemical language model. Consistent, standardized modular searches have yielded molecules with high structural similarity but cannot provide functionally similar yet structurally dissimilar molecules. Using inconsistently standardized molecular SMILES model inputs, the exemplary system was observed to identify molecules that are functionally similar to the query and was also indicated by the associated patent literature to be structurally distinct from the query and thus would have been unlikely to be found with traditional chemical similarity search methods. The current similarity searches use structural elements only to conduct chemical searches. The exemplary method can aid in the discovery of novel structural classes of molecules that achieve target functionality.

Description

CHEMICAL FUNCTIONAL HOMOLOGUE SEARCH USINGINCONSISTENT CANONICALIZED QUERYCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of priority to U.S. Provisional Application No. 63 / 502,392, filed May 15, 2023, which is incorporated by reference herein in its entirety.BACKGROUND

[0002] Chemical similarity searches are widely used in-silico methods for identifying new drug-like molecules. These methods generally rely on structure-based comparisons (e.g., 2D and / or 3D structure-based comparisons) to compute molecular similarity.

[0003] There is a benefit to improving chemical similarity searches.SUMMARY

[0004] A system and method for providing a chemical similarity search of functional homologs using inconsistently standardized molecular SMILES model inputs or other molecular model inputs to a chemical language model. Consistent, standardized modular searches have yielded molecules with high structural similarity but cannot provide functionally similar yet structurally dissimilar molecules. Using inconsistently standardized molecular SMILES model inputs, the exemplary system was observed to identify molecules that are functionally similar to the query and was also indicated by the associated patent literature to be structurally distinct from the query and thus would have been unlikely to be found with traditional chemical similarity search methods. The current similarity searches use structural elements only to conduct chemical searches. The exemplary method can aid in the discovery of novel structural classes of molecules that achieve target functionality.

[0005] The exemplary system employs a chemical language model to generate rich, highdimensional vectors for the vector-based chemical search. The exemplary system utilizes two different chemical string representation algorithms: one for the query and the other for the database. The inconsistently standardized molecular SMILES model inputs, via the prompts, then provide confounding inputs to the model so that when the database is queried, it results in a similarity score based not on molecular features or chemical organization but on the more complex properties of molecular nature learned by the model.

[0006] As used herein, the term “inconsistent standardized” can refer to a different canonicalization resulting from (i) the same canonicalization rules but having another initialization point or (ii) alternative canonicalization rules.

[0007] In an aspect, a method is disclosed comprising providing an input query for a chemical similarity search to a chemical language model that uses language model-encoded feature vectors to compute structural similarity values across molecular space, the chemical language model comprising a neural network with model-encoded feature vectors trained with a SMILES string training data set having a plurality of SMILES strings generated from a first set of canonicalization rules that provide a first standardized consistent grammatical representation of each of the SMILES strings; converting the input query to a SMILES string to second standardized consistent grammatical representation that is inconsistent with the first set of canonicalization rules; searching SMILES strings with the second standardized consistent grammatical representation across the SMILES strings with the first standardized consistent grammatical representation to force identifications of SMILES strings with functional homologues of the input query; and outputting the identified SMILES strings with the functional homologues of the input query.

[0008] In some embodiments, the input query is inconsistently generated from a second set of canonicalization rules different from the first canonicalization rules to provide the second standardized, consistent grammatical representation that is inconsistent with the first set of canonicalization rules.

[0009] In some embodiments, the first canonicalization rules have a corresponding first initialization vector, wherein the input query is inconsistently generated from the first set of canonicalization rules having a second initialization vector different from the first initialization vector to provide the second standardized, consistent grammatical representation that is inconsistent with the first set of canonicalization rules.

[0010] In some embodiments, the chemical language model is a Bidirectional Encoder Representations from Transformers (BERT) model.

[0011] In some embodiments, the SMILES string training data set includes database feature vectors generated for a plurality of non-redundant SMILES strings.

[0012] In some embodiments, the SMILES string training data set and the query SMILES string are normalized to chunks of pre-defined length SMILES string-feature vector pairs.

[0013] In some embodiments, the pre-defined length SMILES string feature vector pairs have a normalized unit length according to the Euclidean norm.

[0014] In some embodiments, the search is performed based on a cosine similarity metric.

[0015] In another aspect, a system comprising a molecule similarity searching engine configured to receive at least one input query to perform a chemical similarity search in a chemical language model that uses language model-encoded feature vectors to compute structural similarity values across molecular space, the chemical language model comprising a neural network with model-encoded feature vectors trained with a SMILES string training data set having SMILES strings generated from a first set of canonicalization rules that provide a first standardized consistent grammatical representation of each of the SMILES strings; and a query interface configured, by instructions, to: receive an input query; convert the input query to a SMILES string to second standardized consistent grammatical representation that is inconsistent with the first set of canonicalization rules; the molecule similarity searching engine being configured to (i) receive he query SMILES string and search the query SMILES strings with the second standardized consistent grammatical representation across the SMILES strings with the first standardized consistent grammatical representation to force identifications of SMILES strings with functional homologues of the input query and (ii) output the identified SMILES strings with the functional homologues of the input query.

[0016] In some embodiments, the input query is inconsistently generated from a second set of canonicalization rules different from the first canonicalization rules to provide the second standardized, consistent grammatical representation that is inconsistent with the first set of canonicalization rules.

[0017] In some embodiments, the first canonicalization rules have a corresponding first initialization vector, wherein the input query is inconsistently generated from the first set of canonicalization rules having a second initialization vector different from the first initialization vector to provide the second standardized, consistent grammatical representation that is inconsistent with the first set of canonicalization rules.

[0018] In some embodiments, the chemical language model is a Bidirectional Encoder Representations from Transformers (BERT) model.

[0019] In some embodiments, the SMILES string training data set includes database feature vectors generated for a plurality of non-redundant SMILES strings.

[0020] In some embodiments, the SMILES string training data set and the query SMILES string are normalized to chunks of pre-defined length SMILES string-feature vector pairs.

[0021] In some embodiments, the pre-defined length SMILES string feature vector pairs have a normalized unit length according to the Euclidean norm.

[0022] In some embodiments, the search is performed based on a cosine similarity metric.

[0023] In another aspect, a non-transitory computer readable medium is disclosed having instructions stored thereon, wherein execution of the instructions by a processor causes the processor to receive an input query for a chemical similarity search to a chemical language model that uses language model-encoded feature vectors to compute structural similarity values across molecular space, the chemical language model comprising a neural network with model-encoded feature vectors trained with a SMILES string training data set having a plurality of SMILES strings generated from a first set of canonicalization rules that provide a first standardized consistent grammatical representation of each of the SMILES strings; convert the input query to a SMILES string to second standardized consistent grammatical representation that is inconsistent with the first set of canonicalization rules; search SMILES strings with the second standardized consistent grammatical representation across the SMILES strings with the first standardized consistent grammatical representation to force identifications of SMILES strings with functional homologues of the input query; and output the identified SMILES strings with the functional homologues of the input query.BRIEF DESCRIPTION OF DRAWINGS

[0024] FIGURE 1 depicts an example method to provide a chemical similarity search of functional homologs using inconsistently standardized molecular model inputs or other molecular model inputs to a chemical language mode.

[0025] FIGURE 2A depicts that the query molecule and chemical database were converted into SMILES strings, canonicalized, and then inputted into a language model to obtain embeddings. The cosine similarity between the query embedding and database embeddings was computed, resulting in a vector of embedding similarities. The database was canonicalized withRDKitAtomO, whereas the queries were canonicalized using one of the following: RDKit AtomO, RDKitAtomn, orOEChem2.3.0. FIGURE 2B depicts feature cosine similarity of each RDKit canonicalized query depending on the chosen root atom number. In order from left to right and top to bottom: zidovudine, penicillin, nirmatrelvir, LSD, fentanyl, SB-759335-B, BMS-536924, 558441-90-0, fluticasone furoate. The canonicalized variant with the lowest feature cosine similarity to the Atom 0 representation was chosen as the “RDKit Atom n” query. The root atoms providing most dissimilar feature vectors were 15 for zidovudine, 13 for penicillin, 21 for nirmatrelvir, 11 for LSD, 17 for fentanyl, 14 for SB-759335-B, 17 for BMS-536924, 10 for 558441-90-0, and 31 for fluticasone furoate. FIGURE 2C depicts the structures of query molecules. All queries were made achiral during canonicalization. From left to right and top to bottom: Zidovudine,Penicillin, Nirmatrelvir, LSD, Fentanyl, SB-759335-B, BMS-536924, 558441-90-0, Fluticasone furoate. FIGURE 2D depicts fingerprint Tanimoto coefficients between each of the query molecules.

[0026] FIGURE 3 depicts similarity metrics between the three canonicalized representations of each query molecule. String similarity, as measured with gestalt pattern matching, demonstrates that different canonicalizations result in markedly different strings. Shared token ratio and token length ratio indicate that these strings were tokenized into different inputs to the CLM. Feature similarity demonstrates that the differently canonicalized queries’ token vectors were interpreted differently by the model resulting in increased spread across feature space. Feature similarity was determined by cosine similarity between ChemBERTa vector embeddings. Deviations from 1.0 for each metric represent divergence between canonicalized queries.

[0027] FIGURES 4A-4D depict results showing search behavior being dependent on canonicalization. FIG. 4A shows similarity metrics for all CheSS searches between each canonicalized query and its respective top 250 results. Compared to consistent canonicalization, the top 250 results from alternative canonicalizations were significantly more dissimilar in structural, scaffold, string, and shared token similarity. Structural similarity was measured by whole-molecule fingerprint Tanimoto similarity, scaffold similarity was measured by scaffold fingerprint Tanimoto similarity, and string similarity was measured by gestalt pattern matching. Asterisks indicate the level of statistical significance for two-sided independent t-tests (ns, p < 1.0; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001). FIGS. 4B-4D show the index rank of each canonicalization’ s top 250 results for zidovudine compared to the index rank that these same molecules scored in a fingerprint Tanimoto search. Black dot indicates molecules functionally similar to the query, as determined by the LLM-assisted patent search. FIGURE 4E depicts the index rank of each canonicalization’ s top 250 results for zidovudine and penicillin compared to the index rank that these same molecules scored in a fingerprint Tanimoto search. Black dot indicates molecules functionally similar to the query, as determined by the GPT-assisted patent search. Top row: Zidovudine. Bottom row: Penicillin. FIGURE 4F depicts the index rank of each canonicalization’ s top 250 results for nirmatrelvir and LSD compared to the index rank that these same molecules scored in a fingerprint Tanimoto search. Black dot indicates molecules functionally similar to the query, as determined by the GPT-assisted patent search. Top row: Nirmatrelvir. Bottom row: LSD. FIGURE 4G depicts the index rank of each canonicalization’ s top 250 results for fentanyl and SB-759335-B compared to the index rankthat these same molecules scored in a fingerprint Tanimoto search. Black dot indicates molecules functionally similar to the query, as determined by the GPT-assisted patent search. Top row: Fentanyl. Bottom row: SB-759335-B. FIGURE 4H depicts the index rank of each canonicalization’s top 250 results for BMS-536924 and 558441-90-0 compared to the index rank that these same molecules scored in a fingerprint Tanimoto search. Black dot indicates molecules functionally similar to the query, as determined by the GPT-assisted patent search. Top row: BMS-536924. Bottom row: 558441-90-0. Figure 41 depicts the index rank of each canonicalization’s top 250 results for fluticasone furoate compared to the index rank that these same molecules scored in a fingerprint Tanimoto search. Black dot indicates molecules functionally similar to the query, as determined by the GPTassisted patent search.

[0028] FIGURES 5A-5M depict a comprehensive list of all molecules considered and their associated patents.

[0029] FIGURES 6A-6D depict patent-derived functional analysis for each canonicalization’s results. FIG. 6A shows mean drug-like & dye-like molecules returned in the top 20 results from drug-like queries (95% CI, n=8). FIG. 6B shows mean drug-like & dye-like molecules returned in the top 20 results from drug-like queries (95% CI, n=8). FIG. 6C shows structure-function categorization across all queries for each canonicalization (n=8 for each canonicalization). The structural similarity was determined by fingerprint Tanimoto similarity (+ indicates Tc> 0.60, and - indicates Tc< 0.60). Functional similarity was determined by patent search (+ indicates similar function, - indicates no known relevant function to query). Criteria for similar function for each molecule were as follows: Penicillin G: antibiotic [3]; nirmatrelvir: protease inhibitor or antiviral

[0036] ; zidovudine: antiviral

[0037] ; LSD: 5-HT receptor agonist or dopaminergic agonist [38, 39]; fentanyl: opioid analgesic or muscarinic receptor agonist [40-42]; acid blue 25 FA: dye or electroluminescent; avobenzone: UV-Absorption, electroluminescent [43, 44]; 2-dPAC: electroluminescent

[0045] . FIG. 6D shows mean non-derivative functional analogues returned in the top 20 results (95% CI, n=8). Asterisks indicate the level of statistical significance for two-sided independent t- tests (ns, P<1.0; *, P<0.05; **, P<0.01; ***, P<0.001; **** P<0.0001).

[0030] FIGURES 7A-7N depict the structures of molecules discussed herein. FIG. 7A shows RD Kit Atom n zidovudine top result #2. FIG. 7B shows RD Kit Atom n LSD top result #7. FIG. 7C shows RDKit Atom n acid blue 25 FA top result #9. FIG. 7D shows RD Kit Atom n avobenzone top result #13. FIG. 7E shows RDKit Atom n fentanyl top result #20. FIG. 7F shows RDKit Atom n penicillin G top result #9. FIG. 7G shows OEChem nirmatrelvir top result #16. FIG. 7H shows OEChem nirmatrelvir top result #17. FIG. 71shows OEChem nirmatrelvir top result #8. FIG. 7J shows OEChem LSD top result #8. FIG. 7K shows OEChem LSD top result #14. FIG. 7L shows OEChem acid blue 25 FA top result #5. FIG. 7M shows OEChem acid blue 25 FA top result #20. FIG. 7N shows OEChem 2- dPAC top result #15.

[0031] FIGURE 8A depicts patent summarization and functional determination pipeline. Chemical SMILES strings were converted to PubChem CIDs based on same connectivity, then associated patent IDs were obtained. These were then used to obtain the patent title, abstract, and description from Google Scholar. The patent information was then passed into GPT-3.5-turbo with a prompt to obtain summarized functional labels. These labels were then passed into GPT-4 with pre-defined labels to determine functional similarity between the query and molecule in question. FIGURE 8B depicts structural and scaffold similarity between each query and its functionally similar search results. The functionally similar molecules, as determined from the LLM-assisted patent search, from the alternate canonicalization search results contained significantly more dissimilar structures and scaffolds compared to when the canonicalization was the same as the database. Structural similarity was measured by whole-molecule fingerprint Tanimoto similarity; scaffold similarity was measured by scaffold fingerprint Tanimoto similarity. Asterisks indicate the level of statistical significance for two-sided independent t-tests (ns, p < 1.0; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001). FIGURES 8C-8E depict that CheSS queries with alternative canonicalization identify functionally similar molecules with divergent structure. Results from the OEChem-canonicalized queries are shown, although RD Kit Atom n also had structurally distinct functional analogues in the top 250 results. FIG. 8C shows Nirmatrelvir, a peptidomimetic SARS-CoV-2 main protease inhibitor, returned a wide range of non-peptidomimetic protease and peptidase inhibitors. FIG. 8D shows LSD, an ergoline 5- HT2A agonist, returned many non-ergoline 5-HT receptor ligands. FIG. 8E shows Fentanyl, an elongated piperidine opioid agonist, returned several morphinan and non-fentanyl opiates.DETAILED DESCRIPTION

[0032] Some references, which may include various patents, patent applications, and publications, are cited in a reference list and discussed in the disclosure provided herein. The citation and / or discussion of such references is provided merely to clarify the description of the present disclosure and is not an admission that any such reference is “prior art” to any aspects of the present disclosure described herein. In terms of notation, “[n] ” corresponds to the nlhreference in the list. All references cited and discussed in this specification areincorporated herein by reference in their entirety and to the same extent as if each reference was individually incorporated by reference.

[0033] Example Methods

[0034] FIG. 1 shows an example method 100 to provide a chemical similarity search of functional homologs using inconsistently standardized molecular model inputs or other molecular model inputs to a chemical language model. Method 100 includes providing (102) an input query for a chemical similarity search to a chemical language model. The chemical language model comprises a neural network with model-encoded feature vectors trained with a molecular model training data set having a plurality of molecular model data (e.g., SMILES strings) generated from a first set of canonicalization rules that provide a first standardized, consistent grammatical representation of each of the molecular model data.

[0035] The input query is inconsistently generated from a second set of canonicalization rules different from the first canonicalization rules to provide the second standardized, consistent grammatical representation that is inconsistent with the first set of canonicalization rules. In some embodiments, the first canonicalization rules have a corresponding first initialization vector, and the input query is inconsistently generated from the first set of canonicalization rules having a second initialization vector different from the first initialization vector.

[0036] In some embodiments, the first canonicalization rules and the second canonicalization rules are different.

[0037] Method 100 further includes converting (104) the input query to a molecular model data to a second standardized, consistent grammatical representation that is inconsistent with the first set of canonicalization rules. In some embodiments, this operation may entail using the same conversion operation as used in the training but with a different initialization vector.

[0038] Method 100 further includes searching (106) the molecular model with the second standardized, consistent grammatical representation across the molecular model data with the first standardized, consistent grammatical representation to force identifications of molecular data with functional homologs of the input query.

[0039] Method 100 further includes outputting (108) the identified molecular model data with the functional homologs of the input query.

[0040] Experimental Results and Additional Examples

[0041] A study was conducted to develop a chemical similarity search pipeline utilizing a transformer encoder-based chemical language model to generate embeddings upon whichsimilarity scores can be computed. From this, the study designed a prompt engineering strategy that expands upon existing chemical semantic searches by creating a method able to identify structurally dissimilar molecules with similar functions. The study demonstrated the utility of this search method to identify non-obvious functional compounds related to multiple different query molecules. This method may aid in repurposing known compounds or in discovering new structural classes of molecules that have desirable functionality. Despite potential drawbacks, the study found that CheSS and the canonicalization prompt engineering method discussed herein have broad interest to the chemical community, as they teach exemplary machine learning operations that can be used outside of staid similarity queries.

[0042] CheSS Overview. The Chemical Semantic Search (CheSS) is a molecular search framework that uses language model-encoded feature vectors to compute similarity scores across molecular space. A database of molecules is encoded as strings using SMILES format

[0017] . A chemical language model is then used to generate a feature vector for each molecule in the database as well as the query molecule. The cosine similarity between the query vector and each database vector is computed, resulting in a vector of feature cosine similarities.

[0043] Language Model. ChemBERTa was used as the language model to generate embeddings

[0029] . This was a Bidirectional Encoder Representations from Transformers (BERT) model with 12 hidden layers of size 768 and was trained on 10M random non- redundant achiral SMILES strings selected from PubChem

[0030] . ChemBERTa was chosen over newer, higher-parameter BERT models due to the ease of implementation and publicly available dataset. ChemBERTa does not support isomeric SMILES (chirality), and all SMILES were canonicalized before input.

[0044] Database. The CheSS molecular feature database was built from the ~10M random achiral molecules used to train ChemBERTa

[0029] . For each molecule, the SMILES string was canonicalized using RD Kit

[0031] . The study reduced the dataset to exclude all SMILES strings that tokenized to more than 512 tokens, the maximum supported by ChemBERTa. This resulted in a database of 9,999,809 molecules. The database SMILES strings were encoded into feature vectors with ChemBERTa. The [CLS] token vector representations of the final layer were chosen to be the feature vectors, as described in the original BERT paper

[0030] . These feature vectors were then L2 normalized and stored in chunks of 100k SMILES string-feature vector pairs for future cosine similarity calculations.

[0045] Canonicalization Query Types. ChemBERTa was trained on SMILES strings canonicalized using RD Kit

[0029] ,

[0031] . Different canonicalization algorithms result in different but equally valid standardized strings representing the same molecule, which the studyutilized to create three highly different queries for the same molecule. The first query type used RD Kit with its default Python implementation settings. This algorithm was used to canonicalize the database & train the model. When converting molecules to SMILES, RDKit allows specification of which atom number to root the SMILES string to. The default is Atom 0, and each atom results in a different representation. The feature cosine similarity was calculated between the default RDKit SMILES and the “Atom n” RDKit SMILES for each atom in the query molecule, as demonstrated in FIG. 2B.

[0046] From these, the study took the most dissimilar “Atom n” SMILES strings to be the second query type for each molecule. To obtain a third dissimilar SMILES representation, OEChem 2.3.0, a markedly different canonicalization algorithm than RDKit, was used

[0032] . These SMILES strings were obtained from the PubChem website.

[0047] Similarity Metrics. Various similarity metrics were used throughout, which include feature cosine similarity, Gestalt pattern matching similarity, fingerprint Tanimoto similarity, token vector length similarity, and token similarity

[0033] ,

[0034] .

[0048] Feature cosine similarity is a distance metric that calculates the angle between two vectors A and B, per Equation 1.(Eq. 1)

[0049] In Equation 1, a cosine similarity of “1” indicates the normalized vectors are the same, “0” means they are orthogonal to one another, and “-1” means they are opposite of one another.

[0050] Gestalt pattern matching was chosen to calculate string similarity. The metric can be calculated by dividing twice the number of matching characters (Km) by the total number of characters in both strings (Si, S2) per Equation 2.Gestalt Similarity y(Eq. 2)

[0051] Matching characters were identified first from the longest common substring, with recursive counts in non-matching regions on both sides of the substring. The metric ranged from a perfect match of “1” to a completely dissimilar string of “0.” The study used the difflib Python implementation of the Gestalt pattern matching algorithm to calculate Gestalt similarity.

[0052] Fingerprint Tanimoto similarity was used to calculate the structural similarity between pairs of molecules. This method encodes substructures into a binary vector and then calculates the Tanimoto similarity between these encoded vectors. The Tanimoto / Jaccard similarity is the number of shared elements (intersection) between two sets A and B, over the total number of unique elements in both sets (union) (Equation 3). Tanimoto Similarity(Eq. 3)

[0053] In Equation 3, the metric ranged from “1 'hall elements shared) to “0” (no elements shared). The RD Kit default implementation of fingerprint Tanimoto similarity was used herein. When fingerprint Tanimoto similarity was performed on the whole molecule, it was occasionally referred to in the paper as structural similarity, and when only performed on the scaffold, it was referred to as scaffold similarity. Scaffolds were obtained using the Murcko scaffold method in RD Kit

[0054] All SMILES were encoded into token vectors before being passed into the model. These tokenized vectors were used for additional comparisons to better understand search behavior. The first metric used from these was the ratio of token lengths between two vectors. The second metric was the token Tanimoto / Jaccard similarity (Equation 3) between the two molecules’ token vectors and was used to determine the ratio of shared tokens between the two vectors. This metric ranges from 1 (all tokens shared) to 0 (no tokens shared).

[0055] LLM-Assisted Patent Summarization and. Functional Similarity Determination. Chemical functionality was derived from the patent literature. To rapidly parse the patent literature to allow for a comprehensive analysis, an LLM-assisted patent summarization pipeline was developed. Using the PubChem database, achrival SMILES strings were converted to their corresponding same-connectivity PubChemCIDs (if available), which were then converted into the set of associated patents. To avoid excess LLM-associated API costs, a limit of 10 patent IDs per molecule was imposed, randomly selected from the set of linked patent IDs.

[0056] The patent IDs were then used as queries to obtain the patent title, abstract, and description from Google Scholar. The descriptions were capped at 3,500 characters to avoid token overflows and excess API costs. These were then appended to the following user prompt: “Return a short set of three 1-3 word descriptors that best describe the chemical or pharmacological function(s) of the molecule described by the given patent title, abstract, and partial description (giving more weight to title and abstract). Be specific and concise, but notnecessarily comprehensive (choose a small number of great descriptor). Follow the syntax ‘{descriptor_l } / {descriptor_2} / {etc}’, writing ‘NA’ if nothing is provided. DO NOT BREAK THIS SYNTAX. The following is the patent”:,” and inputted into OpenAI’s GPT- 3.5-turbo (version accessed on July 15, 2023) with 0 temperature and the system prompt “You are an organic chemist summarizing chemical patents.” GPT-3.5-turbo was chosen over GPT-4 for the patent summarization step due to the high API cost per token for the latter model.

[0057] The functional labels determined from GPT-3.5-turbo were converted into a set, passed into GPT-4 alongside the pre-determined query functional labels, and appended to the prompt, “The following is a set of functional labels for two different molecules. Determine if the two molecules have similar functionality. If similar elements are found in both lists, these molecules have similar functions. You must respond in the format ‘{yes or no} — {20-word maximum explanation}’” with 0 temperature and the system prompt “You are an organic chemistry expert.”

[0058] A random 100 molecules from the 9,750 CheSS search results were used to validate the patent summarization and functional similarity determination pipeline. First, the functional labels were consolidated into a set with their associated patents to assess GPT-3.5- turbo’s ability to summarize patents into quality functional descriptors. The study examined the 320 patents associated with the 100 random molecules and determined the following metrics for each label: label relevant to patent; label refers to the molecule of interest (MOI), a target of MOI, or the downstream effects of MOI; label refers to MOI, the target of MOI, downstream effects of MOI, or molecules of which MOI is an intermediate. These 100 molecules were then used to assess GPT-4’ s ability to determine the functional similarity between the MOI and its associated CheSS query. The CheSS query descriptors were derived from the literature, aiming to be as specific as possible while taking into account the inherent vagueness of patents. For each molecule, the study assessed GPT-4’ s success in determining functional similarity and the validity of its provided reasoning. Accuracy, precision, and recall were calculated for the former.

[0059] Results of the Study

[0060] In silica drug discovery methods have long relied on chemical similarity searches in pharmaceutical pipelines. Recently, language models have been successful in biochemical prediction tasks for chemical properties such as drug-likeness, protein-ligand interactions, and other metrics. As CLMs have been used for the prediction of a multitude of chemical characteristics, it is plausible that embeddings from these models are structured in a way thatapproximates a summary of molecular properties for a given molecule. SMILES representations can be used to represent chemical structures as strings and can be used as inputs to CLMs, but due to the nature of chemical connectivity, there are often many valid representations for the same molecule. This multiplicity of input formats results in the CLM generating different embeddings for the same molecule, which can either be mitigated through string standardization (known as canonicalization) or utilized as a data augmentation technique to create a SMILES representation-invariant CLM. The latter is predominantly used during model training, reducing overfitting to particular strings by teaching the model that a given structure has multiple corresponding string representations.

[0061] However, SMILES augmentation during model inference, rather than model training, may prove to be advantageous for a chemical similarity search. Chemical databases often include structural permutations of well-studied chemicals, causing embedding-based similarity searches to be dominated by low-level structure-based comparisons. SMILES strings representing the same structure often share characters and substrings with one another, the primary difference being the specific canonicalization rules used to assemble the string for a given molecule. Lor CLMs trained on one canonicalization, inputs using unseen canonicalizations would be out of distribution, not on the token identity level but on the immediate token relationship level for a given molecule. This is because the underlying tokens and token relationships are not entirely unknown to the model but are novel in the context of the given molecule. It seems plausible that a vector comparison between embeddings from two different canonicalizations could ignore the differences in low-level string and structural data and instead use whole-molecule properties approximating molecular function, thus serving as a prompt engineering strategy for the discovery of structurally distinct functional analogues.

[0062] To test the hypothesis, the study built a CLM-embedding-based chemical similarity search utilizing one canonicalization for the CLM and database and another canonicalization for the query SMILES strings. A pipeline was developed to perform a chemical similarity search using cosine similarity on CLM embeddings obtained from ChemBERTa, an unsupervised transformer encoder-based model

[0029] . This pipeline was named the Chemical Semantic Search (CheSS), outlined in FIG. 2A.

[0063] The study trained ChemBERTa on SMILES strings canonicalized using the default RD Kit implementation, herein referred to as “RD Kit Atom 0” [29, 31]. The study converted ChemBERTa’ s training set into a molecular database of ~10M RD Kit Atom 0 SMILES strings, upon which CheSS searches were performed. The study then explored howthe CheSS search results were impacted by three different SMILES canonicalizations. The first was RD Kit’s default canonicalization, “RD Kit Atom 0” (TABLE 1). The second canonicalization was created by maximizing the distance of the created string embedding to RD Kit Atom 0 by varying the root atom number, herein referred to as “RD Kit Atom n” (TABLE 1; FIG. 2B). The third used OEChem 2.3.0, a markedly different algorithm, to canonicalize the query and is referred to as “OEChem” (TABLE 1). CheSS searches utilizing these three canonicalizations differed only in the representation of the query molecule, with the database and model remaining constant.TABLE 1. Different canonical SMILES string representations for each query molecule.

[0064] A CheSS search using each of the three query canonicalizations was conducted on nine structurally and functionally distinct molecules: zidovudine, penicillin, nirmatrelvir, lysergic acid diethylamide (LSD), fentanyl, SB-759335-B, BMS-536924, 558441-90-0, and fluticasone furoate (FIGS. 2C-2D).

[0065] Determining Whether Canonicalization Impacts Search Behavior. The study conducted a CheSS search using each of three different query canonicalizations on eight molecules of known function and roughly equal chemical complexity: penicillin G, nirmatrelvir, zidovudine, lysergic acid diethylamide (LSD), fentanyl, acid blue 25 free acid (acid blue 25 FA), avobenzone, and 2-diphenylaminocarbazole (2-dPAC) (FIGS. 2C-2D).

[0066] The study chose these molecules as they are of roughly equal complexity but otherwise represent two distinct classes of molecules: drug-like bioactive molecules and non- drug-like photochemical molecules (herein referred to as dye-like). In addition, the molecules were all sufficiently structurally dissimilar from one another, as determined by having a fingerprint Tanimoto coefficient (Tc) less than 0.60 (FIG. 2D). That said, acid blue 25 FA was more similar to the drug-like molecules, whereas the dyes avobenzone and 2-dPAC were both highly dissimilar from all other query molecules (FIG. 2D).

[0067] Statistical Analysis of Query Canonicalizations. For each query molecule, several similarity metrics were calculated between the three canonicalizations (pairwise comparisons, FIG. 3). Gestalt pattern matching, a string similarity metric, showed that each query canonicalized into different strings, with a mean pairwise value of 0.47 across canonicalizations (n=8) (FIG. 3). Because the CLM does not directly receive strings as inputs but instead receives the tokenized representations (integer-mapped subsections) of strings, thetoken vectors were analyzed to understand how these strings would be presented to the model. Tanimoto similarity, a metric comparing shared elements between two sets, was applied to the token vectors, which indicated that the query strings were converted using markedly different input tokens (mean pairwise value of 0.69 across canonicalizations) (FIG. 3). Similarly, token vector lengths were variable in length, with some queries differing by almost a factor of 2 depending on canonicalization (FIG. 3). Changes in the token vectors cause differences in featurization, or the model’s interpretation of said input, and it was found that different embeddings were obtained depending on canonicalization (mean pairwise feature cosine similarity of 0.66), indicating that the model interpreted different canonicalizations of the same molecule as quite distinct inputs (FIG. 3).

[0068] String, Token, and Structural Relevance of Top Search Results. To explore how different canonicalizations affect vector search behavior, similarity metrics were obtained comparing each canonicalized query to its respective top 250 CheSS search results. Queries canonicalized with RD Kit Atom 0 yielded compounds high in structural similarity, as measured by the fingerprint Tanimoto coefficient (a metric of molecular substructure similarity), which had a mean coefficient of 0.67 (n = 2,250) (FIG. 4A). In contrast, the mean fingerprint Tanimoto coefficients for RDKit Atom n and OEChem were 0.45 and 0.40, respectively. Similarly, the Murcko scaffold fingerprint coefficient followed this trend with a mean scaffold similarity of 0.64, 0.33, and 0.30 for RDKit Atom 0, RDKit Atom n, and OEChem. For RDKit Atom 0 canonicalized queries, 27.5% of the top 250 results would be identified using a fingerprint Tanimoto search with a cutoff as high as 0.80, indicating that nearly a quarter of the results were 1-2 atomic changes away from the query molecule (FIGS. 4A-4B). In contrast, only 2.8% and 1.8% of the top results for RDKit Atom n and OEChem would have been found from this same high-cutoff fingerprint search. This indicates a significant structural divergence in results when using alternative query canonicalizations (FIGS. 4A-4D).

[0069] It was found that RDKit Atom 0 canonicalized queries returned molecules with a high number of shared tokens to the query token vector (mean ratio of 0.74), whereas this was not the case with RDKit Atom n and OEChem (means of 0.57 and 0.58). This indicates that the model’s ability to memorize tokens to determine embedding similarity was reduced by querying with different canonicalizations (FIG. 4A). Interestingly, it was observed that the token vector length ratios for all canonicalizations’ results had mean values around 1.0, indicating that token vector length heavily influences feature space location and thus similarity (FIG. 4A). This means that token vector length may constrain CLM -basedsimilarity searches to confined regions of chemical space, with alternative canonicalizations acting as ways to bypass this predominant search criterion through variations in token vector length, thereby allowing exploration of distant regions of chemical space to create more comprehensive and far-reaching similarity searches (FIGS. 4C-4D).

[0070] Patent & Literature Search. In order to determine the known functionality of the molecules examined herein, a patent & literature search was performed. The list of patents and literature articles about the molecules was obtained from PubChem

[0035] . A comprehensive list of all molecules considered herein and their associated patents is provided in FIG. 5.

[0071] Patent Search Reveals Functionality of Molecules. In order to begin to determine the functional significance of the search results, patent and literature searches were conducted on the top 20 results from each search. In general, functionally drug-like queries returned high levels of drug-like molecules and few dye-like molecules, and conversely, queries on functionally dye-like molecules returned more dye-like molecules and many fewer drug-like molecules (FIGS. 6A-6B).

[0072] An exception to the latter statement were molecules identified using OEChem inquiries, but this skew was due almost solely to results from the avobenzone search (which in turn has known biological activity [46, 47]). In general, the baseline of random drug-like molecules returned for drug-based queries exists but is relatively low.

[0073] The functional similarity of query results was contrasted with fingerprint Tanimoto similarity (FIG. 5C). The categorization of functionality was either positive (known relevant functionality to the query) or negative (unknown relevant functionality), and the structural similarity was either positive (Tc> 0.60) or negative (Tc< 0.60).Criteria / ontologies for similar functionality for each molecule were as follows: Penicillin G: antibiotic [3]; nirmatrelvir: protease inhibitor or antiviral

[0036] ; zidovudine: antiviral

[0037] ; LSD: 5-HT receptor agonist or dopaminergic agonist [38, 39]; fentanyl: opioid analgesic or muscarinic receptor agonist [40-42]; acid blue 25 FA: dye or electroluminescent; avobenzone: UV-Absorption, electroluminescent [43, 44]; 2-dPAC: electroluminescent

[0045] . To illustrate, for OEChem-canonicalized nirmatrelvir (a SARS-CoV-2 main protease inhibitor), several top results (7, 8, 16, and 17) were classified as positive, as these compounds were known protease inhibitors and / or antivirals (FIGS. 7H-7J). A table of the top 20 results for each search, complete with links to their PubChem pages, relevant patents, and functional descriptors, is listed in FIG. 5.

[0074] As expected, CheSS searches with the query molecules canonicalized with RDKit Atom 0 resulted in the identification of molecules with similar structures and functions. There were 31 structurally dissimilar molecules (Tc< 0.60) with shared functionality to the query, 30 of which were nonetheless structural derivatives. Penicillin G returned P-lactam antibiotics, nirmatrelvir returned Hepatitis C Virus (HCV) protease inhibitors, zidovudine returned antiviral pyrimidine nucleosides, LSD returned psychoactive ergolines, fentanyl returned narcotic piperidine analogues, acid blue 25 FA returned anthraquinone dyes, avobenzone returned dibenzoylmethane permutants, and 2-dPAC returned electroluminescent carbazole and triphenylamine derivatives.

[0075] In contrast to RDKit Atom 0 queries, RDKit Atom n queries returned molecules that had greater structural diversity but that still contained many structural analogues. There were 26 structurally dissimilar molecules (Tc< 0.60) with shared functionality to the query, some 22 of which were structural derivatives, including purine and pyrimidine analogs of zidovudine (FIG. 7A). However, these hits also included a quite distinct dopaminergic agonist for LSD (FIG. 7B), a hydrophobic fluorescence probe for acid blue 25 FA (FIG. 7C), a refractive copolymer for avobenzone (FIG. 7D), and a fentanyl-like agonist for its known coronary muscarinic receptor (FIG. 7E) [41, 42]. 2-dPAC did not return any molecules with known relevant functionality using this canonicalization, though the results had similar aromaticity to the query. Interestingly, tetracycline antibiotics have been used to inhibit metalloproteases, and the penicillin G query returned two non-tetracycline metalloprotease inhibitors (FIG. 7F)

[0048] .

[0076] OEChem generally returned molecules that were structurally highly dissimilar to the query. There were 35 structurally dissimilar molecules (Tc< 0.60) with shared functionality to the query, and in contrast to the previous two canonicalizations, only 7 of these were structural derivatives. The more diverse compounds with shared functionality included two non-HCV protease inhibitors (FIGS. 7G-7H); a Respiratory Syncytial Virus antiviral for nirmatrelvir (FIG. 71); a dopaminergic and serotonergic agonist (FIG. 7 J) and a 5-HT1 receptor agonist for LSD (FIG. 6K); porphyrins (FIG. 6L) and other conjugated dyes for acid blue 25 FA (FIG. 6M); photovoltaic and electroluminescent molecules for avobenzone; and highly conjugated electroluminescent molecules for 2-dPAC (FIG. 6N). Taken together, these hits provide anecdotal proof for the hypothesis that changing query canonicalizations can lead to the discovery of novel, but functional, chemical compounds. These insights may provide interesting avenues for drug repurposing, and molecules with previously unknown functions may serve as leads for novel drug discovery.

[0077] Large language model-assisted patent search reveals functional analogues. The data discussed thus far demonstrates that alternatively canonicalized queries find structurally distinct compounds but do not confirm whether the returned molecules are functionally relevant to the query. Patent literature represents a rich source of information on molecular functionality. Summarizing patents into brief and relevant functional descriptors is a tedious task that would take an intractable amount of time over the 9,750 CheSS search results’ 17,980 unique associated patents. This limitation was bypassed with the creation of an automated large language model (LLM)-assisted patent summarization and functional determination pipeline.

[0078] This workflow takes a SMILES string, finds PubChem compound IDs (CIDs) with the same connectivity, obtains linked patent IDs (limited to 10 per molecule), and scrapes Google Scholar to obtain the patent title, abstract, and description (FIG. 8A). This information is then combined with a prompt and passed into GPT-3.5-turbo to obtain a set of brief one- to three-word functional descriptors (FIG. 8A). The descriptors are aggregated per molecule and combined into a second prompt alongside pre-defined descriptors of the relevant query molecule’s functionality, which is then passed into GPT-4 to determine whether the two molecules have similar functionality (FIG. 8A). This method was manually validated for performance on a random 100 molecules from the 9,750 search results, analyzing both label accuracy per patent (n = 1,028 labels) and functional similarity to the relevant query (n = 100 molecules) (TABLE 2, TABLE 3, TABLE 4). The results of this validation demonstrated 95% accuracy, 94% precision, and 96% recall on functional similarity determination, with the limiting factor being that some of the molecules had been linked to patents in which they served as intermediates rather than the patent product itself.TABLE 2. Validation for LLM-assisted patent summarization. Achiral SMILES for the MOI were converted to same-connectivity PubChem CIDs, and associated patents were retrieved and fed into GPT-3.5-turbo for summarization into brief labels. Manual validation was performed on 100 molecules randomly selected from the 27 CheSS searches’ top 250 results.TABLE 3. Validation for LLM-assisted functional similarity. Patent summarization labels were fed into GPT-4 alongside labels for the query, and the model was prompted to determine if the two sets of labels describe molecules with similar functionality. Manual validation was performed on 100 molecules randomly selected from the 27 CheSS searches’ top 250 results.TABLE 4. Functional descriptors of query molecules. Functional descriptors were initially chosen to represent well-documented specific protein targets (e.g., reverse transcriptase inhibitor, AKT inhibitor). Vague patents necessitated the use of broader functional descriptors, including specific protein target classes (e.g., protease inhibitor, serotonergic) and downstream effects (e.g., analgesic, antiviral).

[0079] Functional similarities of the top 250 results for their respective nine query molecules were analyzed (FIG. 8B). Seventy-five percent of the RD Kit Atom 0 results were functionally similar to their query, 42% for RD Kit Atom n, and 27% for OEChem. However, the alternative canonicalizations returned molecules that contained significantly dissimilar structures and scaffolds from the query, with mean whole-molecule fingerprint Tanimotosimilarities of 0.72, 0.54, and 0.51 for RDKit Atom 0, RDKit Atom n, and OEChem, respectively, and mean scaffold fingerprint Tanimoto similarities of 0.69, 0.43, and 0.41 (FIG. 8A).

[0080] Upon closer inspection of the structurally dissimilar functional analogues from the alternatively canonicalized CheSS queries, it was found that many of these molecules represented structural classes distinct from the query. These are illustrated with the nirmatrelvir, LSD, and fentanyl OEChem-canonicalized searches (FIGS. 8C-8E).Nirmatrelvir, a peptidomimetic severe acute respiratory syndrome coronavirus 2 (SARS- CoV-2) main protease inhibitor, returned a wide range of non-peptidomimetic protease and peptidase inhibitors (FIG. 8C). LSD, an ergoline 5-HT2A agonist, returned many non- ergoline 5 -hydroxy tryptamine (5-HT, serotonin) receptor ligands not obviously derivable from LSD (FIG. 8D). Fentanyl, an elongated piperidine opioid agonist, returned several morphinan and non-fentanyl opiates (FIG. 8E).

[0081] One artifact of determining functional similarity from patent literature is that the true positive rate for functionally relevant molecules may be higher than the value obtained in this study, as the relevant assay is not guaranteed to have been conducted and reported for a given molecule. As a consequence, the molecules deemed to have irrelevant functionality are curiously the most desirable leads for novel drug discovery and repurposing. Taken together, these hits provide support for the hypothesis that changing query canonicalization for a transformer-based chemical similarity search can lead to the discovery of structurally novel yet functionally relevant chemical compounds.

[0082] Explanation of Search Behavior. The study found that alternative canonicalizations influence the search behavior of CheSS through changes in tokenization, which causes the CLM to weight higher-order relationships, more importantly when creating embeddings. There are stark differences between RDKit and OEChem canonicalizations, notably their differences in the representation of aromatic rings. OEChem prefers the Kekule form (C1=CC=CC=C1), while RDKit prefers to use lowercase with assumed aromaticity (clcccccl), and these differences, among others, result in markedly different tokenization, both in the composition and the length of the tokenized vectors (FIG. 3).

[0083] SMILES strings generated from alternative canonicalizations can be viewed as out-of-distribution queries, not in the token identities (as no new tokens were created by changing canonicalization) but in the low-level token relationships for a given molecule. This relaxation of embedding the immediate token relationships can be seen in the diminished shared token ratios of alternatively canonicalized queries’ results as low-level patternmatching can no longer occur (FIG. 4A). However, higher-order token relationships are still maintained across all searches regardless of canonicalization, as demonstrated by the constraint on token vector length (FIG. 4A). It is the other opaque high-order token relationships that is believed to be causing the language model to embed functionally similar molecules, independent of structure, close together in feature space when alternative canonicalizations are used (FIG. 8A).

[0084] When CLMs are forced to go beyond simple token patterns to determine similarity, more nuanced relationships may appear. Given the nature of transformers, it is indiscernible what these relationships are, but based on the analysis, the study found it possible that the CLM may, for example, key on the apposition of functional groups in space, in a way similar to how receptors perceive ligands.

[0085] Example Trained Machine Learning Model

[0086] Machine Learning. In addition to the machine learning model described above, the analysis system can be implemented using one or more artificial intelligence and machine learning operations. The term “artificial intelligence” can include any technique that enables one or more computing devices or computing systems (i.e., a machine) to mimic human intelligence. Artificial intelligence (Al) includes but is not limited to knowledge bases, machine learning, representation learning, and deep learning. The term “machine learning” is defined herein to be a subset of Al that enables a machine to acquire knowledge by extracting patterns from raw data. Machine learning techniques include, but are not limited to, logistic regression, support vector machines (SVMs), decision trees, Naive Bayes classifiers, and artificial neural networks. The term “representation learning” is defined herein to be a subset of machine learning that enables a machine to automatically discover representations needed for feature detection, prediction, or classification from raw data. Representation learning techniques include, but are not limited to, autoencoders and embeddings. The term “deep learning” is defined herein to be a subset of machine learning that enables a machine to automatically discover representations needed for feature detection, prediction, classification, etc., using layers of processing. Deep learning techniques include but are not limited to artificial neural networks or multilayer perceptron (MLP).

[0087] Machine learning models include supervised, semi- supervised, and unsupervised learning models. In a supervised learning model, the model learns a function that maps an input (also known as feature or features) to an output (also known as target) during training with a labeled data set (or dataset). In an unsupervised learning model, the algorithm discovers patterns among data. In a semi-supervised model, the model learns a function thatmaps an input (also known as a feature or features) to an output (also known as a target) during training with both labeled and unlabeled data.

[0088] Neural Networks. An artificial neural network (ANN) is a computing system including a plurality of interconnected neurons (e.g., also referred to as “nodes”). This disclosure contemplates that the nodes can be implemented using a computing device (e.g., a processing unit and memory as described herein). The nodes can be arranged in a plurality of layers, such as an input layer, an output layer, and optionally one or more hidden layers with different activation functions. An ANN having hidden layers can be referred to as a deep neural network or multilayer perceptron (MLP). Each node is connected to one or more other nodes in the ANN. For example, each layer is made of a plurality of nodes, where each node is connected to all nodes in the previous layer. The nodes in a given layer are not interconnected with one another, i.e., the nodes in a given layer function independently of one another. As used herein, nodes in the input layer receive data from outside of the ANN, nodes in the hidden layer(s) modify the data between the input and output layers, and nodes in the output layer provide the results. Each node is configured to receive an input, implement an activation function (e.g., binary step, linear, sigmoid, tanh, or rectified linear unit (ReLU) function), and provide an output in accordance with the activation function. Additionally, each node is associated with a respective weight. ANNs are trained with a dataset to maximize or minimize an objective function. In some implementations, the objective function is a cost function, which is a measure of the ANN’s performance (e.g., error such as LI or L2 loss) during training, and the training algorithm tunes the node weights and / or bias to minimize the cost function. This disclosure contemplates that any algorithm that finds the maximum or minimum of the objective function can be used for training the ANN. Training algorithms for ANNs include but are not limited to backpropagation. It should be understood that an artificial neural network is provided only as an example machine learning model. This disclosure contemplates that the machine learning model can be any supervised learning model, semi- supervised learning model, or unsupervised learning model. Optionally, the machine learning model is a deep learning model. Machine learning models are known in the art and are therefore not described in further detail herein.

[0089] A convolutional neural network (CNN) is a type of deep neural network that has been applied, for example, to image analysis applications. Unlike traditional neural networks, each layer in a CNN has a plurality of nodes arranged in three dimensions (width, height, depth). CNNs can include different types of layers, e.g., convolutional, pooling, and fully- connected (also referred to herein as “dense”) layers. A convolutional layer includes a set offilters and performs the bulk of the computations. A pooling layer is optionally inserted between convolutional layers to reduce the computational power and / or control overfitting (e.g., by downsampling). A fully-connected layer includes neurons, where each neuron is connected to all of the neurons in the previous layer. The layers are stacked similarly to traditional neural networks. GCNNs are CNNs that have been adapted to work on structured datasets such as graphs.

[0090] Other Supervised Learning Models. A logistic regression (LR) classifier is a supervised classification model that uses the logistic function to predict the probability of a target, which can be used for classification. LR classifiers are trained with a data set (also referred to herein as a “dataset”) to maximize or minimize an objective function, for example, a measure of the LR classifier’s performance (e.g., an error such as LI or L2 loss), during training. This disclosure contemplates that any algorithm that finds the minimum of the cost function can be used. LR classifiers are known in the art and are therefore not described in further detail herein.

[0091] An Naive Bayes’ (NB) classifier is a supervised classification model that is based on Bayes’ Theorem, which assumes independence among features (i.e., the presence of one feature in a class is unrelated to the presence of any other features). NB classifiers are trained with a data set by computing the conditional probability distribution of each feature given a label and applying Bayes’ Theorem to compute the conditional probability distribution of a label given an observation. NB classifiers are known in the art and are therefore not described in further detail herein.

[0092] A k-NN classifier is an unsupervised classification model that classifies new data points based on similarity measures (e.g., distance functions). The k-NN classifiers are trained with a data set (also referred to herein as a “dataset”) to maximize or minimize a measure of the k-NN classifier’s performance during training. This disclosure contemplates any algorithm that finds the maximum or minimum. The k-NN classifiers are known in the art and are therefore not described in further detail herein.

[0093] A majority voting ensemble is a meta-classifier that combines a plurality of machine learning classifiers for classification via majority voting. In other words, the majority voting ensemble’s final prediction (e.g., class label) is the one predicted most frequently by the member classification models. The majority voting ensembles are known in the art and are therefore not described in further detail herein.

[0094] Example Computing System

[0095] It should be appreciated that the logical operations described above and in the appendix can be implemented (1) as a sequence of computer- implemented acts or program modules running on a computing system and / or (2) as interconnected machine logic circuits or circuit modules within the computing system. The implementation is a matter of choice dependent on the performance and other requirements of the computing system. Accordingly, the logical operations described herein are referred to variously as state operations, acts, or modules. These operations, acts, and / or modules can be implemented in software, in firmware, in special purpose digital logic, in hardware, and any combination thereof. It should also be appreciated that more or fewer operations can be performed than shown in the figures and described herein. These operations can also be performed in a different order than those described herein.

[0096] The computer system is capable of executing the software components described herein for the exemplary method or systems. In an embodiment, the computing device may comprise two or more computers in communication with each other that collaborate to perform a task. For example, but not by way of limitation, an application may be partitioned in such a way as to permit concurrent and / or parallel processing of the instructions of the application. Alternatively, the data processed by the application may be partitioned in such a way as to permit concurrent and / or parallel processing of different portions of a data set by the two or more computers. In an embodiment, virtualization software may be employed by the computing device to provide the functionality of a number of servers that are not directly bound to the number of computers in the computing device. For example, virtualization software may provide twenty virtual servers on four physical computers. In an embodiment, the functionality disclosed above may be provided by executing the application and / or applications in a cloud computing environment. Cloud computing may comprise providing computing services via a network connection using dynamically scalable computing resources. Cloud computing may be supported, at least in part, by virtualization software. A cloud computing environment may be established by an enterprise and / or can be hired on an as-needed basis from a third-party provider. Some cloud computing environments may comprise cloud computing resources owned and operated by the enterprise as well as cloud computing resources hired and / or leased from a third-party provider.

[0097] In its most basic configuration, a computing device includes at least one processing unit and system memory. Depending on the exact configuration and type of computing device, system memory may be volatile (such as random-access memory (RAM)),non-volatile (such as read-only memory (ROM), flash memory, etc.), or some combination of the two.

[0098] The processing unit may be a standard programmable processor that performs arithmetic and logic operations necessary for the operation of the computing device. While only one processing unit is shown, multiple processors may be present. As used herein, processing unit and processor refers to a physical hardware device that executes encoded instructions for performing functions on inputs and creating outputs, including, for example, but not limited to, microprocessors (MCUs), microcontrollers, graphical processing units (GPUs), and application- specific circuits (ASICs). Thus, while instructions may be discussed as executed by a processor, the instructions may be executed simultaneously, serially, or otherwise executed by one or multiple processors. The computing device may also include a bus or other communication mechanism for communicating information among various components of the computing device.

[0099] Computing devices may have additional features / functionality. For example, the computing device may include additional storage such as removable storage and nonremovable storage including, but not limited to, magnetic or optical disks or tapes. Computing devices may also contain network connection(s) that allow the device to communicate with other devices, such as over the communication pathways described herein. The network connection(s) may take the form of modems, modem banks, Ethernet cards, universal serial bus (USB) interface cards, serial interfaces, token ring cards, fiber distributed data interface (FDDI) cards, wireless local area network (WLAN) cards, radio transceiver cards such as code division multiple access (CDMA), global system for mobile communications (GSM), long-term evolution (LTE), worldwide interoperability for microwave access (WiMAX), and / or other air interface protocol radio transceiver cards, and other well-known network devices. Computing devices may also have input device(s) such as keyboards, keypads, switches, dials, mice, trackballs, touch screens, voice recognizers, card readers, paper tape readers, or other well-known input devices. Output device(s) such as printers, video monitors, liquid crystal displays (LCDs), touch screen displays, displays, speakers, etc., may also be included. The additional devices may be connected to the bus in order to facilitate the communication of data among the components of the computing device. All these devices are well known in the art and need not be discussed at length here.

[0100] The processing unit may be configured to execute program code encoded in tangible, computer-readable media. Tangible, computer-readable media refers to any media that is capable of providing data that causes the computing device (i.e., a machine) to operatein a particular fashion. Various computer-readable media may be utilized to provide instructions to the processing unit for execution. Example tangible, computer-readable media may include but is not limited to volatile media, non-volatile media, removable media, and non-removable media implemented in any method or technology for storage of information such as computer-readable instructions, data structures, program modules, or other data. System memory, removable storage, and non-removable storage are all examples of tangible computer storage media. Example tangible, computer-readable recording media include, but are not limited to, an integrated circuit (e.g., field-programmable gate array or applicationspecific IC), a hard disk, an optical disk, a magneto-optical disk, a floppy disk, a magnetic tape, a holographic storage medium, a solid-state device, RAM, ROM, electrically erasable program read-only memory (EEPROM), flash memory or other memory technology, CD- ROM, digital versatile disks (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices.

[0101] In light of the above, it should be appreciated that many types of physical transformations take place in the computer architecture in order to store and execute the software components presented herein. It also should be appreciated that the computer architecture may include other types of computing devices, including hand-held computers, embedded computer systems, personal digital assistants, and other types of computing devices known to those skilled in the art.

[0102] In an example implementation, the processing unit may execute program code stored in the system memory. For example, the bus may carry data to the system memory, from which the processing unit receives and executes instructions. The data received by the system memory may optionally be stored on the removable storage or the non-removable storage before or after execution by the processing unit.

[0103] It should be understood that the various techniques described herein may be implemented in connection with hardware or software or, where appropriate, with a combination thereof. Thus, the methods and apparatuses of the presently disclosed subject matter, or certain aspects or portions thereof, may take the form of program code (i.e., instructions) embodied in tangible media, such as floppy diskettes, CD-ROMs, hard drives, or any other machine-readable storage medium wherein, when the program code is loaded into and executed by a machine, such as a computing device, the machine becomes an apparatus for practicing the presently disclosed subject matter. In the case of program code execution on programmable computers, the computing device generally includes a processor, a storage medium readable by the processor (including volatile and non-volatile memory and / orstorage elements), at least one input device, and at least one output device. One or more programs may implement or utilize the processes described in connection with the presently disclosed subject matter, e.g., through the use of an application programming interface (API), reusable controls, or the like. Such programs may be implemented in a high-level procedural or object-oriented programming language to communicate with a computer system. However, the program(s) can be implemented in assembly or machine language, if desired. In any case, the language may be a compiled or interpreted language, and it may be combined with hardware implementations.

[0104] Discussion

[0105] Applications for small molecules in modem society are various and widespread, including treatment of heritable disease, pathogen inhibition, and the generation of functional materials for use in electronics and consumer goods. Molecular function emerges from structure, but it is not always obvious how this emerges from first principles due to the dependence of function on the target molecule [1]. Traditionally, exploration of natural products has led to the identification of vital pharmaceuticals and specialty chemicals [2-5]. These first generation molecules act as starting points, upon which new molecules are engineered for furthering desired functionality [6]. Structural neighbors often share similar functionality, as the relevant chemistry may be unchanged or improved [7]. However, molecules with low structural similarity can act on the same target despite the highly different structure, as is the case with morphine and fentanyl on the p-opioid receptor [8].

[0106] There are numerous contemporary approaches to applying machine learning to chemistry [9-16]. However, the application of language models to this space has led to surprising success in predicting biochemical features such as drug-likeness and protein-ligand interactions

[0011] ,

[0013] . These methods require string representations of molecules, commonly using the Simplified Molecular- Input Line-Entry System (SMILES)

[0017] . Language models are often trained in an unsupervised manner with the reward function tied to sequence reconstruction, i.e., feeding the model a masked or partial input with the goal of reconstructing the original sequence. It was recently demonstrated that a chemical language model, though trained only on SMILES strings, correctly predicted complex biophysical and quantum-chemical properties

[0018] . This points to the possibility that these models develop a chemical latent space that allows for the emergence of higher-order biochemical comprehension.

[0107] Recently, computationally generated chemical libraries have grown to surpass 37 billion commercially available compounds

[0019] . This marked growth has generated a newfield of computational pre-screening of chemicals in order to support resource-efficient discovery in the laboratory

[0020] . One primary class of computational pre-screening methods is chemical similarity searches. These methods have historically used structure-based comparisons, a notable example being the fingerprint Tanimoto search which computes a hierarchical list of molecules ranked by molecular substructure similarity to a given query [21, 22].

[0108] Chemical Language Models (CLM) have been applied to drug discovery, in particular, de novo molecule generation and chemical similarity searches. De novo methods generate novel molecules with the decoder portion of a language model after fine-tuning toward a specific molecule or target

[0014] —

[0016] ,

[0018] . Building off the recent success of generative models such as GPT, de novo molecule generation has shown promise but is limited due to a lack of generalizability and guaranteed synthesizability

[0023] ,

[0024] . In contrast, a chemical similarity search based on a CLM has the advantage of computational speed, generalizability, and high database control to ensure synthesizability. Sellner et al. recently created a novel transformer-based chemical similarity search with an optimized loss function to approximate previous structure-based methods

[0012] . However, this method tends to identify molecules with high structural similarity to the query molecule, when instead, the study would often like to find structurally distinct functional analogues. Such a CLM -based search does not currently exist.

[0109] Here, the study described a chemical similarity search based on a CLM that identifies molecules with similar function to a given query molecule. This method works by calculating the CLM-computed feature vector similarity between a query SMILES string and a chemical database. Keeping the SMILES canonicalization algorithm constant between the query and database resulted in a chemical language search that approximated recent transformer-based chemical search methods. However, the study found that when the query SMILES string was canonicalized with a different algorithm than was used for the database, the reliance on structural similarity diminished while functional similarity was retained. This behavior seemed reasonable given the literature describing how models learn to better represent chemical space when SMILES randomization occurs during training and from the reported emergent understanding of underrepresented languages in predominantly English- trained models [25-28]. The study utilized alternatively canonicalized queries as a novel prompt engineering strategy to identify structurally distinct functional analogues of small molecules. The exemplary method fundamentally differs from existing literature in that SMILES augmentation was used for the query of a chemical similarity search rather than formodel training and fine-tuning to a specific task. The study tested the exemplary method across three canonicalizations and eight query molecules and found that with increasingly divergent canonicalizations, the study was able to identify structurally distinct functional analogues.

[0110] Chemical similarity searches historically relied upon substructure searches that apply a one-hot encoding based on the presence or absence of a structural feature. Natural language processing (NLP) algorithms have allowed for the creation of models that can generate rich vector representations of molecules. By using these vectors, the exemplary can conduct a similarity search based on these rich vectors. In addition, the exemplary system employs a molecular search framework that uses language model-encoded feature vectors to compute similarity scores across molecular space. A database of molecules can be encoded as strings using SMILES format. A chemical language model is used to generate a feature vector for each molecule in the database and the query molecule. The cosine similarity between the query vector and each database vector is computed, resulting in a vector of feature cosine similarities. Canonicalization is an algorithmic approach used to standardize a given molecule’s SMILES string to a consistent representation among all valid possibilities. By standardizing a query molecule using an alternative algorithm to that used in database generation, the resulting cosine similarity search identifies molecules with diverse structures yet similar functions. This can effectively be used to search chemical space for functional homologues.

[0111] The unusual method applied in this case is the use of an alternative canonicalization algorithm used to query the database. In doing so, the exemplary system was observed to identify molecules that are functionally similar to the query in addition to being structurally different. The identified molecules have been observed to be structurally distinct from the query and associatively different from those considered to be determined via traditional chemical similarity search methods. The exemplary method thus can aid in the discovery of novel structural classes of molecules that achieve target functionality.

[0112] Conclusion

[0113] Each and every feature described herein, and each and every combination of two or more of such features, is included within the scope of the present invention, provided that the features included in such a combination are not mutually inconsistent.

[0114] Although example embodiments of the disclosed technology are explained in detail herein, it is to be understood that other embodiments are contemplated. Accordingly, it is not intended that the disclosed technology be limited in its scope to the details ofconstruction and arrangement of components set forth in the following description or illustrated in the drawings. The disclosed technology is capable of other embodiments and of being practiced or carried out in various ways.

[0115] It must also be noted that, as used in the specification and the appended claims, the singular forms “a,” “an” and “the” include plural referents unless the context clearly dictates otherwise. Ranges may be expressed herein as from “about” or “approximately” one particular value and / or to “about” or “approximately” another particular value. When such a range is expressed, other exemplary embodiments include from the one particular value and / or to the other particular value.

[0116] By ‘ ‘comprising” or “containing” or “including” is meant that at least the named compound, element, particle, or method step is present in the composition or article or method, but does not exclude the presence of other compounds, materials, particles, method steps, even if the other such compounds, material, particles, method steps have the same function as what is named.

[0117] Unless otherwise expressly stated, it is in no way intended that any method set forth herein be construed as requiring that its steps be performed in a specific order. Accordingly, where a method claim does not actually recite an order to be followed by its steps or it is not otherwise specifically stated in the claims or descriptions that the steps are to be limited to a specific order, it is no way intended that an order be inferred, in any respect. This holds for any possible non-express basis for interpretation, including: matters of logic with respect to arrangement of steps or operational flow; plain meaning derived from grammatical organization or punctuation; the number or type of embodiments described in the specification.

[0118] While the methods and systems have been described in connection with certain embodiments and specific examples, it is not intended that the scope be limited to the particular embodiments set forth, as the embodiments herein are intended in all respects to be illustrative rather than restrictive.

[0119] The following patents, applications, and publications, as listed below and throughout this document, are hereby incorporated by reference in their entirety herein. References[1] Qingxin Li and CongBao Kang. Mechanisms of action for small molecules revealed by structural biology in drug discovery. International Journal of Molecular Sciences, 21(15):5262, 2020.[2] Gordon M Cragg and David J Newman. Biodiversity: A continuing source of novel drug leads. Pure and applied chemistry, 77(l):7-24, 2005.[3] Alexander Fleming. Penicillin. British medical journal, 2(4210):386, 1941.[4] Rui H Jiao, Shu Xu, Jun Y Liu, Hui M Ge, Hui Ding, Chen Xu, Hai L Zhu, and Ren XTan. Chaetominine, a cytotoxic alkaloid produced by endophytic chaetomium sp. ifb- e015. Organic Letters, 8(25):5709-5712, 2006.[5] Mansukh C Wani and Susan Band Horwitz. Nature as a remarkable chemist: A personal story of the discovery and development of taxol®. Anti-cancer drugs, 25(5):482, 2014.[6] James P Hughes, Stephen Rees, S Barrett Kalindjian, and Karen L Philpott. Principles of early drug discovery. British journal of pharmacology, 162(6): 1239-1249, 2011.[7] Yvonne C Martin, James L Kofron, and Linda M Traphagen. Do structurally similar molecules have similar biological activity? Journal of medicinal chemistry, 45(19):4350-4358, 2002.[8] Hasan Pathan and John Williams. Basic opioid pharmacology: an update. British journal of pain, 6(1): 11-16, 2012.[9] Limeng Pu, Misagh Naderi, Tairan Liu, Hsiao-ChunWu, Supratik Mukhopadhyay, andMichal Brylinski. etoxpred: a machine learning-based approach to estimate the toxicity of drug candidates. BMC Pharmacology and Toxicology, 20(1): 1-15, 2019.

[0010] Yilin Yang, Mingjie Liu, and John R Kitchin. Neural network embeddings based similarity search method for atomistic systems. Digital Discovery, 1 (5):636— 644, 2022.

[0011] Kyunghoon Lee, Jinho Jang, Seonghwan Seo, Jaechang Lim, and Woo Youn Kim. Drug-likeness scoring based on unsupervised learning. Chemical Science, 13(2):554— 565, 2022.

[0012] Manuel S Sellner, Amr H Mahmoud, and Markus A Lili. Efficient virtual high-content screening using a distance-aware transformer model. Journal of Cheminformatics, 15(1):18, 2023.

[0013] Bomin Wei, Yue Zhang, and Xiang Gong. Deeplpi: a novel deep learning-based model for protein-ligand interaction prediction for drug repurposing. Scientific reports, 12(1): 18200, 2022.

[0014] Michael Moret, Francesca Grisoni, Paul Katzberger, and Gisbert Schneider. Perplexitybased molecule ranking and bias estimation of chemical language models. Journal of chemical information and modeling, 62(5): 1199-1206, 2022.

[0015] Michael Moret, Irene Pachon Angona, Leandro Cotos, Shen Yan, Kenneth Atz, Cyrill Brunner, Martin Baumgartner, Francesca Grisoni, and Gisbert Schneider. Leveragingmolecular structure and bioactivity with chemical language models for de novo drug design. Nature Communications, 14(1): 114, 2023.

[0016] Daniel Flam-Shepherd, Kevin Zhu, and Alan Aspuru-Guzik. Language models can learn complex molecular distributions. Nature Communications, 13( 1):3293, 2022.

[0017] David Weininger. Smiles, a chemical language and information system. 1. introduction to methodology and encoding rules. Journal of chemical information and computer sciences, 28(1):31— 36, 1988.

[0018] Jerret Ross, Brian Belgodere, Vijil Chenthamarakshan, Inkit Padhi, Youssef Mroueh, and Pay el Das. Large-scale chemical language representations capture molecular structure and properties. Nature Machine Intelligence, 4(12): 1256-1264, 2022.

[0019] Benjamin I Tingle, Khanh G Tang, Mar Castanon, John J Gutierrez, Munkhzul Khurelbaatar, Chinzorig Dandarchuluun, Yurii S Moroz, and John J Irwin. Zinc-22- a free multi-billion-scale database of tangible compounds for ligand discovery. Journal of Chemical Information and Modeling, 63(4): 1166-1176, 2023.

[0020] Maria Batool, Bilal Ahmad, and Sangdun Choi. A structure-based drug discovery paradigm. International journal of molecular sciences, 20(11):2783, 2019.

[0021] Katalin Szilagyi, Beata Flachner, Istvan Hajdu, Maria Szaszko, Krisztina Dobi, Zsolt L'orincz, Sandor Cseh, and Gybrgy Dorman. Rapid identification of potential drug candidates from multi-million compounds’ repositories, combination of 2d similarity search with 3d ligand / structure based methods and in vitro screening. Molecules, 26 (18):5593, 2021.

[0022] Dagmar Stumpfe and Jurgen Bajorath. Similarity searching. Wiley Interdisciplinary Reviews: Computational Molecular Science, l(2):260-282, 2011.

[0023] Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners. Advances in neural information processing systems, 33:1877-1901, 2020.

[0024] Wenhao Gao and Connor W Coley. The synthesizability of molecules proposed by generative models. Journal of chemical information and modeling, 60(12):5714-5723, 2020.

[0025] Josep Arus-Pous, Simon Viet Johansson, Oleksii Prykhodko, Esben Jannik Bjerrum, Christian Tyrchan, Jean-Louis Reymond, Hongming Chen, and Ola Engkvist. Randomized smiles strings improve the quality of molecular generative models. Journal of cheminformatics, 11(1): 1-13, 2019.

[0026] Maranga Mokaya, Fergus Imrie, Willem P van Hoorn, Aleksandra Kalisz, Anthony R Bradley, and Charlotte M Deane. Testing the limits of smiles-based de novo molecular generation with curriculum and deep reinforcement learning. Nature Machine Intelligence, pages 1-9, 2023.

[0027] Jason Wei, Yi Tay, Rishi Bommasani, Colin Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama, Maarten Bosma, Denny Zhou, Donald Metzler, et al. Emergent abilities of large language models. arXiv preprint arXiv:2206.07682, 2022.

[0028] Freda Shi, Mirac Suzgun, Markus Freitag, Xuezhi Wang, Suraj Srivats, Soroush Vosoughi, Hyung Won Chung, Yi Tay, Sebastian Ruder, Denny Zhou, et al. Language models are multilingual chain-of-thought reasoners. arXiv preprint arXiv:2210.03057, 2022.

[0029] Seyone Chithrananda, Gabriel Grand, and Bharath Ramsundar. Chemberta: Large-scale self-supervised pretraining for molecular property prediction. CoRR, abs / 2010.09885, 2020..

[0030] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: pretraining of deep bidirectional transformers for language understanding. CoRR, abs / 1810.04805, 2018. 04805.

[0031] Greg Landrum et al. Rdkit: A software suite for cheminformatics, computational chemistry, and predictive modeling. Greg Landrum, 8, 2013.

[0032] TK OEChem. version 2.3.0; openeye scientific software: Santa fe, nm.

[0033] John W Ratcliff and David E Metzener. Pattem-matching-the gestalt approach. Dr Dobbs Journal, 13(7):46, 1988.

[0034] David Bajusz, Anita Racz, and Karoly Heberger. Why is tanimoto index an appropriate choice for fingerprint-based similarity calculations? Journal of cheminformatics, 7(1): 1— 13, 2015.

[0035] Sunghwan Kim, Paul A Thiessen, Evan E Bolton, Jie Chen, Gang Fu, Asta Gindulyte, Lianyi Han, Jane He, Siqian He, Benjamin A Shoemaker, et al. Pubchem substance and compound databases. Nucleic acids research, 44(D1): D1202-D1213, 2016.

[0036] Jennifer Hammond, Heidi Leister-Tebbe, Annie Gardner, Paula Abreu, Weihang Bao, Wayne Wisemandle, MaryLynn Baniecki, Victoria M Hendrick, Bharat Damle, Abraham Simon-Campos, et al. Oral nirmatrelvir for high-risk, nonhospitalized adults with covid- 19. New England Journal of Medicine, 386(15): 1397-1408, 2022.

[0037] Margaret A Fischl, Douglas D Richman, Nellie Hansen, Ann C Collier, John T Carey, Michael F Para, W David Hardy, Raphael Dolin, William G Powderly, J Davis Allan, etal. The safety and efficacy of zidovudine (azt) in the treatment of subjects with mildly symptomatic human immunodeficiency virus type 1 (hiv) infection: a double-blind, placebo-controlled trial. Annals of internal medicine, 112(10):727-737, 1990.

[0038] M Titeler, RA Lyon, and RA Glennon. Radioligand binding evidence implicates the brain 5-ht 2 receptor as a site of action for Isd and phenylisopropylamine hallucinogens. Psychopharmacology, 94:213-216, 1988.

[0039] Lorenzo Pieri, Margherita Pieri, and Willy Haefely. Lsd as an agonist of dopamine receptors in the striatum. Nature, 252:586-588, 1974.

[0040] Patil Armenian, Kathy T Vo, Jill Barr-Walker, and Kara L Lynch. Fentanyl, fentanyl analogs and novel synthetic opioids: a comprehensive review. Neuropharmacology, 134:121-132, 2018.

[0041] Takao Yamanoue, Jose M Brum, Fawzy G Estafanous, Philip A Khairallah, and Carlos M Ferrario. Fentanyl attenuates porcine coronary arterial contraction through m3- muscarinic antagonism. Anesthesia & Analgesia, 76 (2):382-390, 1993.

[0042] Olav Hustveit. Binding of fentanyl and pethidine to muscarinic receptors in rat brain. The Japanese Journal of Pharmacology, 64(1):57— 59, 1994.

[0043] S Afonso, K Horita, JP Sousa e Silva, IF Almeida, MH Amaral, PA Lobao, PC Costa, Margarida S Miranda, Joaquim CG Esteves da Silva, and JM Sousa Lobo. Photodegradation of avobenzone: Stabilization effect of antioxidants. Journal of Photochemistry and Photobiology B: Biology, 140:36-40, 2014.

[0044] Yuyang Zhou, Jinyong Zhuang, Wenming Su, and Xiaomei Wang. Yellow organic lightemitting diodes from heteroleptic iridium (iii) complexes with avobenzone ligands as dopants. European Journal of Inorganic Chemistry, 2015(33):5571-5576, 2015.

[0045] A Venkateswararao, KR Justin Thomas, Chuan-Pei Lee, and Kuo-Chuan Ho. Effect of auxiliary chromophores on the optical, electrochemical, and photovoltaic properties of carbazole-based dyes. Asian Journal of Organic Chemistry, 4( 1):69— 80, 2015.

[0046] Sungjin Ahn, Seungchan An, Moonyoung Lee, Eunyoung Lee, Jeong Joo Pyo, Jeong Hyeon Kim, Min Won Ki, Sun Hee Jin, Jaehyoun Ha, and Minsoo Noh. A long-wave uva filter avobenzone induces obesogenic phenotypes in normal human epidermal keratinocytes and mesenchymal stem cells. Archives of toxicology, 93:1903-1915, 2019.

[0047] Changwon Yang, Whasun Lim, FullerWBazer, and Gwonhwa Song. Avobenzone suppresses proliferative activity of human trophoblast cells and induces apoptosis mediated by mitochondrial disruption. Reproductive Toxicology, 81:50-57, 2018.

[0048] Michele M Castro, Arulmozhi D Kandasamy, Nermeen Youssef, and Richard Schulz. Matrix metalloproteinase inhibitor properties of tetracyclines: therapeutic potential in cardiovascular diseases. Pharmacological Research, 64(6):551-560, 2011.

[0049] Fabio Urbina, Filippa Lentzos, Cedric Invemizzi, and Sean Ekins. Dual use of artificial- intelligence-powered drug discovery. Nature Machine Intelligence, 4(3): 189— 191, 2022.

[0050] Sellner et al. Journal of Cheminformatics (2023) 15:18 DOI: 10.1186 / sl3321-023- 00686-z

[0051] Chithranada et al. arXiv (2020) DOI: 10.48550 / arXiv.2010.09885

[0052] Ross et al. Nature Machine Intelligence (2022) DOI: 10.1038 / s42256-022-00580-7.

Claims

What we claim is:

1. A method comprising: providing an input query for a chemical similarity search to a chemical language model that uses language model-encoded feature vectors to compute structural similarity values across molecular space, the chemical language model comprising a neural network with model-encoded feature vectors trained with a SMILES string training data set having a plurality of SMILES strings generated from a first set of canonicalization rules that provide a first standardized consistent grammatical representation of each of the SMILES strings; converting the input query to a SMILES string to second standardized consistent grammatical representation that is inconsistent with the first set of canonicalization rules; searching SMILES strings with the second standardized consistent grammatical representation across the SMILES strings with the first standardized consistent grammatical representation to force identifications of SMILES strings with functional homologues of the input query; and outputting the identified SMILES strings with the functional homologues of the input query.

2. The method of claim 1, wherein the input query is inconsistently generated from a second set of canonicalization rules different from the first canonicalization rules to provide the second standardized consistent grammatical representation that is inconsistent with the first set of canonicalization rules.

3. The method of claim 1, wherein the first canonicalization rules have a corresponding first initialization vector, and wherein the input query is inconsistently generated from the first set of canonicalization rules having a second initialization vector different from the first initialization vector, to provide the second standardized consistent grammatical representation that is inconsistent with the first set of canonicalization rules.

4. The method of any one of claims 1-3, wherein the chemical language model is a Bidirectional Encoder Representations from Transformers (BERT) model.

5. The method of any one of claims 1-4, wherein the SMILES string training data set includes database feature vectors generated for a plurality of non-redundant SMILES strings.

6. The method of any one of claims 1-5, wherein the SMILES string training data set and the query SMILES string are normalized to chunks of pre-defined length SMILES stringfeature vector pairs.

7. The method of claim 6, wherein the pre-defined length SMILES string feature vector pairs has a normalized unit length according to Euclidean norm.

8. The method of any one of claims 1-7, wherein the searching is preformed based on a cosine similarity metric.

9. A system comprising: a molecule similarity searching engine configured to receive at least one input query to perform a chemical similarity search in a chemical language model that uses language model-encoded feature vectors to compute structural similarity values across molecular space, the chemical language model comprising a neural network with model-encoded feature vectors trained with a SMILES string training data set having SMILES strings generated from a first set of canonicalization rules that provide a first standardized consistent grammatical representation of each of the SMILES strings; and a query interface configured, by instructions, to: receive an input query; convert the input query to a SMILES string to second standardized consistent grammatical representation that is inconsistent with the first set of canonicalization rules; the molecule similarity searching engine being configured to (i) receive the query SMILES string and search the query SMILES strings with the second standardized consistent grammatical representation across the SMILES strings with the first standardized consistent grammatical representation to force identifications of SMILES strings with functional homologues of the input query and (ii) output the identified SMILES strings with the functional homologues of the input query.

10. The system of claim 9, wherein the input query is inconsistently generated from a second set of canonicalization rules different from the first canonicalization rules to provide the second standardized consistent grammatical representation that is inconsistent with the first set of canonicalization rules.

11. The system of claim 9, wherein the first canonicalization rules have a corresponding first initialization vector, and wherein the input query is inconsistently generated from the first set of canonicalization rules having a second initialization vector different from the first initialization vector, to provide the second standardized consistent grammatical representation that is inconsistent with the first set of canonicalization rules12. The system of any one of claims 9-11, wherein the chemical language model is a Bidirectional Encoder Representations from Transformers (BERT) model.

13. The system of any one of claims 9-12, wherein the SMILES string training data set includes database feature vectors generated for a plurality of non-redundant SMILES strings.

14. The system of any one of claims 9-13, wherein the SMILES string training data set and the query SMILES string are normalized to chunks of pre-defined length SMILES stringfeature vector pairs.

15. The system of claim 14, wherein the pre-defined length SMILES string feature vector pairs has a normalized unit length according to Euclidean norm.

16. The system of any one of claims 9-15, wherein the searching is preformed based on a cosine similarity metric.

17. A non-transitory computer readable medium having instructions stored thereon, wherein execution of the instructions by a processor causes the processor to: receive an input query for a chemical similarity search to a chemical language model that uses language model-encoded feature vectors to compute structural similarity values across molecular space, the chemical language model comprising a neural network with model-encoded feature vectors trained with a SMILES string training data set having a plurality of SMILES strings generated from a first set of canonicalization rules that provide a first standardized consistent grammatical representation of each of the SMILES strings; convert the input query to a SMILES string to second standardized consistent grammatical representation that is inconsistent with the first set of canonicalization rules;search SMILES strings with the second standardized consistent grammatical representation across the SMILES strings with the first standardized consistent grammatical representation to force identifications of SMILES strings with functional homologues of the input query; and output the identified SMILES strings with the functional homologues of the input query.

18. A non-transitory computer readable medium of claim 17 having instructions to perform the method of any one of claims 1-8 or execute the system of any one of claims 9-16.

19. A composition having the CID described herein.

Images