A Screening Method for Perfluoroalkyl and Polyfluoroalkyl Substances Based on Molecular Generation

Abstract

The present invention belongs to the field of environmental pollutant screening and relates to a screening method for perfluoroalkyl and polyfluoroalkyl substances based on molecular generation. The present invention utilizes a molecular generation model to expand the existing PFAS structures, generating a large list of potential PFAS structures to overcome the problem of limited coverage of the PFAS chemical space by existing databases, thereby improving the screening throughput of unknown PFAS. The effects and benefits of the present invention are to significantly expand the chemical space coverage of PFAS, achieve efficient screening of unknown PFAS in the environment, increase the screening throughput, and successfully identify a variety of unreported potential PFAS structures, providing effective support for PFAS screening, monitoring, and risk assessment.

Description

Technical Field

[0001] The present invention belongs to the field of environmental pollutant screening and relates to a screening method for perfluoroalkyl and polyfluoroalkyl compounds based on molecular generation. Background Art

[0002] Perfluoroalkyl and polyfluoroalkyl substances (PFAS) are a class of chemicals widely used in industry and commerce. Due to their strong stability, they are difficult to degrade in the environment and are prone to bioaccumulation, posing a potential threat to human and environmental health. Screening PFAS in environmental media helps to assess its potential risks. However, there are tens of thousands of PFAS on the global market, and with the continuous update, use, and release of PFAS, the number of PFAS in the environment is incalculable. Therefore, comprehensively screening PFAS in the environment is a major challenge. Currently, the screening of PFAS mainly relies on high-resolution mass spectrometry technology for non-targeted analysis (CN202411155389.7, September 24, 2024). This technology relies on the existing secondary mass spectrometry of PFAS as reference information to confirm the structure of PFAS in the sample. However, the number of PFAS covered in the reference mass spectrometry library is limited, restricting the comprehensive coverage of the PFAS chemical space by the existing technology and resulting in a low screening throughput.

[0003] To overcome this limitation, some studies have introduced machine learning tools (CN202310655991.6, March 1, 2024; Science Advances, 2024, 10(21), eadn1039) to improve the recognition ability of PFAS by matching experimental mass spectrometry with molecular structures. For example, predict the molecular fingerprint for the secondary mass spectrometry of compounds in the sample and match the fingerprints with similar candidate structures from a pre-obtained PFAS structure database to achieve PFAS structure identification. The PFAS structure database mainly comes from lists organized by international organizations and public databases. For example, the Organization for Economic Cooperation and Development (OECD) PFAS list containing approximately 4,700 PFAS structures and the PubChem database. Although this method has expanded the scope of PFAS recognition to a certain extent, due to the insufficient representativeness of PFAS in the existing database and the fact that many structures are still not publicly available, identifying PFAS structures outside the known database remains a major challenge. Therefore, it is necessary to develop new methods to expand the existing database and increase the coverage of the PFAS chemical space to reveal potential PFAS structures in the environment and improve the PFAS screening throughput.

[0004] Molecular generation models are data-driven artificial intelligence tools with the ability to efficiently explore chemical space and propose potentially effective structures (Nature Machine Intelligence, 2021, 3, (11), 973-984). This technology has been successfully applied in the fields of drug discovery and material design, contributing to the generation of new molecules with specific characteristics. Given that unknown PFAS in the environment may be precursors, by-products, or transformation products of known PFAS, and there are associations in structure or properties between the two, and their distributions in chemical space are close. Molecular generation models can reveal PFAS structures that may appear in the environment by learning the distribution of PFAS structures in the existing inventory in chemical space and generating new structures with similar chemical space distributions. Summary of the Invention

[0005] The technical problem to be solved by the present invention is that traditional PFAS screening methods cover a limited PFAS chemical space and have a low screening throughput. The purpose of the present invention is to provide a PFAS screening method based on a molecular generation model. This method expands the PFAS chemical space covered by existing methods by generating a large number of potential PFAS structures, so as to improve the screening throughput for unknown PFAS.

[0006] Technical Solution of the Present Invention:

[0007] A method for screening perfluoroalkyl and polyfluoroalkyl compounds based on molecular generation, the steps are as follows:

[0008] Step (1) Dataset Preparation

[0009] Using a publicly available PFAS inventory, extract PFAS with clear chemical structures and meeting the OECD definition. The Simplified Molecular Input Line Entry Specification (SMILES) is used to represent the extracted chemical structures, and the RDKit tool is used for the standardization and redundancy removal of SMILES to ensure that the dataset does not contain invalid or duplicate structures. The finally obtained data is used as the training set.

[0010] Step (2) Construction of the Molecular Generation Model

[0011] The long short-term memory network (LSTM) is used as the model architecture, and the model is trained based on the PFAS structures in the training set obtained in step (1). Two strategies are used to make up for the problem of insufficient training data volume, specifically as follows:

[0012] Molecular Generation Model Based on Data Augmentation: Utilize the non-uniqueness of SMILES to increase the diversity of the training set by generating different non-standardized SMILES expressions. Use the SmilesEnumerator program to generate n different SMILES representations for each PFAS structure in the training set and use them as the input for LSTM training. The LSTM architecture based on data augmentation includes an embedding layer of 128 dimensions and a hidden layer of 512 dimensions. During training, the Adam optimizer is used, with a batch size of 128 and a learning rate of 0.001.

[0013] Molecular Generation Model Based on Transfer Learning: First, pre-train the LSTM using a large chemical structure database, and then fine-tune the model on the training set to capture the unique chemical features of PFAS. The LSTM architecture based on transfer learning consists of a total of four layers, with two hidden layers having dimensions of 1024 and 256 respectively, located between two batch normalization layers. The Adam optimizer is used for training, with a batch size of 128 and a learning rate of 0.001. During the fine-tuning process, the parameters of the first hidden layer of the pre-trained model remain unchanged, while the training learning rate of the second layer is reduced to 0.0001.

[0014] Step (3) Molecular Generation

[0015] Use the two trained molecular generation models in step (2) to sample in the PFAS chemical space and combine all the generated SMILES. Use the RDKit tool to validate the validity of the generated SMILES, and then filter them to retain the SMILES structures that conform to the OECD PFAS definition.

[0016] Step (4) Candidate List Construction

[0017] Use the RDKit tool to calculate the molecular formula and exact mass of each generated molecule, count the frequency of each chemical structure being resampled by the model, and integrate the molecular formula, exact mass, and sampling frequency information into a new suspected list for PFAS screening.

[0018] Step (5) PFAS Screening

[0019] Embed the new suspected list obtained in step (4) into the PFAS screening process to screen the high-resolution mass spectrometry data of the samples. Based on specific mass spectrometry features (mass defect > 35, mass defect normalized by carbon number < 0.03, -0.25 < mass normalized by carbon number < 0.1, and Kendrick mass defect > 0.85 or < 0.15), filter the peaks. Finally, retain the peaks with matching of two or more PFAS diagnostic fragments. Extract candidate structures from the new suspected list based on the criteria of an exact mass error less than ±5 ppm and an isotope distribution score greater than 0.8, and sort them in descending order based on their sampling frequencies.

[0020] Advantages of the present invention:

[0021] The present invention significantly expands the PFAS chemical space coverage of the existing database and increases the queryable range of PFAS screening by generating more than 1.4 million potential PFAS structures.

[0022] Based on the generated structures and sampling frequencies provided by the present invention, PFAS screening can effectively prioritize candidate structures, improving the screening efficiency and accuracy of PFAS.

[0023] Based on the strategy provided by the present invention, screening of reported real samples was carried out, and 88 previously unrecognized PFAS were successfully identified, increasing the PFAS identification throughput by 20%. Description of the Drawings

[0024] Figure 1 is the sampling frequency ranking of the correct candidate structures for the retention set match;

[0025] Figure 2 is the screening result of 23 target substances based on the generated structures and sampling frequencies;

[0026] Figure 3 is to identify PFAS in the effluent sample of fluorochemical wastewater based on the new suspected list;

[0027] Figure 4 is to identify PFAS in the influent sample of fluorochemical wastewater based on the new suspected list combined with APP_ID. Detailed Embodiments

[0028] To better understand the content of the present invention, the present invention will be further described below through embodiments.

[0029] Example 1

[0030] The newly constructed PFAS suspected list of the present invention (including structures overlapping with PubChem) is used to retrieve candidate structures for a pre-prepared retention set (containing 408 PFAS structures independent of the training set). Assuming that the exact mass of PFAS in the retention set is the measured exact mass, candidate structures within a relative mass error of 5 ppm are retrieved from the new suspected list, and the candidate structures are prioritized in descending order of sampling frequency. The results show that 91% of the PFAS structures in the retention set can be matched to accurate candidate structures. This indicates that the generative model constructed in the present invention can generate real PFAS structures that do not appear in the training set but can appear in the environment, proving that the PFAS generated by the present invention covers a wide chemical space and has reliable structures. For the accurately matched candidate structures, the proportion of the sampling frequency ranked first among the candidate structures is 56%, the top 5 is 82%, and the top 10 is 90%( Figure 1 ), proving the effectiveness of the strategy of prioritizing candidate structures based on sampling frequency.

[0031] Example 2

[0032] The newly constructed PFAS suspected list of the present invention is used for screening the liquid chromatography-high resolution mass spectrometry data of a mixed solution containing 23 PFAS standards (the PFAS structures covered by the standards do not appear in the training set, and the specific information is shown in Table 1). Through mass spectrometry features, PFAS characteristic peaks are extracted, and based on the exact mass and isotope distribution inferred from their first-order mass spectrometry, candidate structures are retrieved from the new suspected list (the absolute error of the exact mass number is less than 5 ppm, and the isotope distribution score is greater than 0.8), and the candidate structures are prioritized based on their sampling frequency. The results show that among the 23 target PFASs, accurate candidate structures of 20 PFASs are listed as the first candidate molecules (as Figure 2 shown). That is, based on the screening method of the present invention, the screening accuracy of PFAS reaches 87%, proving that the strategy of screening PFAS based on the molecular generative model can accurately identify PFAS in samples.

[0033] Table 1 Information of 23 PFASs

[0034]

[0035]

[0036] Example 3

[0037] The method proposed in the present invention was used for the screening of PFAS in the reported effluent samples of fluorochemical wastewater (Science Advances, 2024, 10(21), eadn1039). The list constructed in the present invention was compared with the currently known largest PFAS structure database, and the structures that overlapped with the known structures in the list were further removed to form a new list for screening. Combining the new list with the PubChem database, the mass spectrometry data of the samples was screened for suspected cases according to the method in Example 2, and a total of 530 potential PFAS features were identified. The results of the match based on the new list were compared with the reported screening results (442 potential PFAS; Science Advances, 2024, 10(21), eadn1039). The results showed that, combined with the PFAS list generated in the present invention, 88 previously unreported potential PFAS structures were successfully identified (such as Figure 3 as shown), and the throughput increased by 20% compared with the original screening results. It was proved that the PFAS screening strategy based on the molecular generation model could improve the PFAS screening throughput and identify PFAS not covered by the known database.

[0038] Example 4

[0039] Combining the new PFAS suspected list proposed in the present invention with the PFAS automatic screening platform APP_ID (Science Advances, 2024, 10(21), eadn1039), the PFAS in the reported influent samples of fluorochemical wastewater was screened. APP_ID extracted the potential PFAS peaks in the mass spectrometry data of the samples through molecular networking, extracted candidate structures from the new list according to the mass error threshold (5 ppm) and the isotope distribution score threshold (0.8), predicted the molecular fingerprints corresponding to the potential PFAS secondary mass spectra, calculated the similarity scores of their molecular fingerprints with the candidate structures, and sorted the candidate structures. The results showed that based on the new PFAS suspected list, a total of 229 potential PFAS were identified (see Figure 4 ), and 115 new potential PFAS were found compared with the original screening results (Science Advances, 2024, 10(21), eadn1039), and the throughput increased by 27%. In addition, new candidate structures were matched for 105 identified potential PFAS, and the best candidate structures for 63 potential PFAS were all newly generated structures, proving that the PFAS screening strategy based on the molecular generation model could supplement the existing PFAS database and provide more reliable candidate structures.

Claims

1. A screening method for perfluoroalkyl and polyfluoroalkyl compounds based on molecular generation, characterized in that, The steps are as follows: Step (1) Dataset preparation Using the publicly available PFAS suspected list, extract PFAS with clear chemical structures that meet the OECD definition; represent the extracted chemical structures in SMILES format, and use the RDKit tool to standardize and remove redundancy of SMILES to ensure that the dataset does not contain invalid or duplicate structures. The finally obtained data is used as the training set; Step (2) Construction of the molecular generation model Adopt the long short-term memory network LSTM as the model architecture, and train the model based on the PFAS structures in the training set obtained in step (1); including two strategies, specifically as follows: Molecular generation model based on data augmentation: Use the SmilesEnumerator program to generate n different SMILES representations for each PFAS structure in the training set, and use them as the input for LSTM training; Molecular generation model based on transfer learning: First pre-train LSTM using a chemical structure database, and then fine-tune it on the training set; Step (3) Molecular generation Use the two trained molecular generation models in step (2) to sample in the PFAS chemical space, and merge all the generated SMILES; Use the RDKit tool to verify the validity of the generated SMILES, and then filter them to retain the SMILES structures that meet the OECD PFAS definition; Step (4) Candidate list construction Use the RDKit tool to calculate the molecular formula and exact mass of each generated molecule, and count the frequency of each chemical structure being repeatedly sampled by the model. Integrate the molecular formula, exact mass, and sampling frequency information into a new suspected list for PFAS screening; Step (5) PFAS screening Embed the new suspected list obtained in step (4) into the PFAS screening process to screen the high-resolution mass spectrometry data of the sample; filter the peaks based on specific mass spectrometry features; finally retain the peaks with two or more PFAS diagnostic fragment matches; Based on the criteria of an exact mass error less than ±5 ppm and an isotope distribution score greater than 0.8, extract candidate structures from the new suspected list and sort them in descending order according to the sampling frequency of the candidate structures.

2. The screening method of perfluoroalkyl and polyfluoroalkyl compounds based on molecular generation according to claim 1, wherein In step (2), the LSTM architecture based on data augmentation includes an embedding layer with 128 dimensions and a hidden layer with 512 dimensions; the Adam optimizer is used during training, the batch size is 128, and the learning rate is 0.

001.

3. The screening method of perfluoroalkyl and polyfluoroalkyl compounds based on molecular generation according to claim 1, characterized in that, In step (2), the LSTM architecture based on transfer learning includes four layers, two of which are hidden layers with dimensions of 1024 and 256 respectively, located between two batch normalization layers; the Adam optimizer is used for training, the batch size is 128, and the learning rate is 0.

001.

4. A screening method for perfluoroalkyl and polyfluoroalkyl compounds based on molecular generation according to claim 1, characterized in that, In the molecular generation model based on transfer learning in step (2), during the fine-tuning process, the parameters of the first layer of the pre-trained model's hidden layer remain unchanged, while the training learning rate of the second layer is reduced to 0.0001.

5. A screening method for perfluoroalkyl and polyfluoroalkyl compounds based on molecular generation according to claim 1, characterized in that, In step (5), the specific mass spectrometry features are: mass defect > 35, mass defect normalized by carbon number < 0.03, -0.25 < mass normalized by carbon number < 0.1, and Kendrick mass defect > 0.85 or < 0.15.

Images