Methods of predicting retention time and retention-related intrinsic parameters for peptide and relevant apparatuses
A neural network model predicts retention times and intrinsic parameters for peptides, addressing the inefficiencies of existing LC optimization methods by providing accurate and efficient peptide separation without extensive experimental data.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- WUXI BIOLOGICS (SHANGHAI) CO LTD
- Filing Date
- 2025-12-31
- Publication Date
- 2026-07-09
AI Technical Summary
Existing methods for optimizing liquid chromatography (LC) parameters for peptide separation, such as one-factor-at-a-time (OFAT) and Design of Experiments (DoE), are labor-intensive and time-consuming, and require significant experimental data, failing to account for complex interactions among variables.
A neural network model is used to predict retention time and intrinsic parameters for peptides by encoding amino acid sequences and LC parameters, allowing for precise prediction of retention times and optimization of LC conditions without extensive experimental data.
The model achieves accurate and efficient prediction of retention times and LC parameter optimization, reducing the need for labor-intensive experiments and enhancing peptide separation efficiency.
Smart Images

Figure CN2025147965_09072026_PF_FP_ABST
Abstract
Description
METHODS OF PREDICTING RETENTION TIME AND RETENTION-RELATED INTRINSIC PARAMETERS FOR PEPTIDE AND RELEVANT APPARATUSESCROSS-REFERENCE TO RELATED APPLICATION
[0001] This application is based on and claims priority to International Application No. PCT / CN2025 / 070777, filed on January 06, 2025, the disclosure of which is incorporated herein by reference in its entirety.TECHNICAL FIELD
[0002] The present disclosure relates to the field of proteomics and biopharmaceutical industry, and more particularly, to methods of predicting a retention time and retention-related intrinsic parameters for a peptide and relevant apparatuses.BACKGROUND
[0003] Peptide separation by liquid chromatography (LC) such as reversed-phase liquid chromatography (RPLC) is a crucial step in peptide mapping, which is a release test to verify the identity and integrity of biotherapeutic proteins through detailed analysis and characterization of peptide mixtures derived from enzymatic digestion.
[0004] Retention and elution are achieved primarily by hydrophobic interactions between the stationary phase in the chromatographic column and peptides, with hydrophobicity modified by the mobile phase. The diverse physicochemical properties of peptides, such as molecular weight, polarity, and charge, contribute to variations in their retention behaviors.
[0005] In order to achieve effective separation, LC parameters such as the type of column, mobile phase composition, pH, additives, temperature, mobile phase gradient, and flow rate need to be meticulously optimized.SUMMARY
[0006] A brief overview of the present disclosure is given below in order to provide a basic understanding on some aspects of the present disclosure. However, it should be understood that this overview is not an exhaustive overview of the present disclosure. It is not intended to identify a key or important part of the present disclosure, nor is it intended to limit the scope of the present disclosure. The purpose thereof is merely to give some concepts of the present disclosure in a simplified form, as a preface to a more detailed description given later.
[0007] According to a first aspect of the present disclosure, a method for predicting intrinsic parameters of a retention theory equation for a peptide is provided. The retention theory equation defines a retention time as a function of LC parameters and the intrinsic parameters. The method includes obtaining an embedding of an amino acid sequence of the peptide, and inputting the embedding to a neural network model trained for predicting the intrinsic parameters of the retention theory equation, so as to obtain the intrinsic parameters for the peptide. The neural network model includes an encoder configured to encode the embedding to generate an encoded embedding, and a parameter prediction network configured to generate the intrinsic parameters for the peptide based on the encoded embedding.
[0008] According to a second aspect of the present disclosure, a method for predicting a retention time for a peptide is provided. The method includes: obtaining intrinsic parameters for the peptide using the method according to the first aspect of the present disclosure; obtaining LC parameters to be used for separation of the peptide; and calculating a retention time for the peptide based on the obtained intrinsic parameters and the obtained LC parameters through the retention theory equation.
[0009] According to a third aspect of the present disclosure, a method for predicting a retention time for a peptide is provided. The method includes: obtaining an embedding of an amino acid sequence of the peptide; obtaining LC parameters to be used for separation of the peptide; and inputting the embedding and the LC parameters to a neural network model trained for predicting the retention time, so as to obtain the retention time for the peptide. The neural network model includes an encoder configured to encode the embedding to generate an encoded embedding, and a prediction head configured to generate the retention time for the peptide based on the encoded embedding and the LC parameters.
[0010] According to a fourth aspect of the present disclosure, A method is provided, which includes at a computer system having a display: receiving a first input containing first preliminary ultra-violet (UV) chromatography data and first mass spectrometry (MS) data of a plurality of peptides separated with a first combination of LC parameters, wherein peak areas and widths and amino acid sequences of the plurality of peptides are identified from the first preliminary UV chromatography data and the first MS data; generating a plurality of combinations of LC parameters each having a default mobile phase gradient, other LC parameters than the mobile phase gradient in each of the plurality of combinations of LC parameters corresponding to a respective point in a parameter design space with each dimension corresponding to a respective one of the other LC parameters; for each of the plurality of combinations of LC parameters, predicting retention times for one or more selected peptides of the plurality of peptides using the method according to the second or third aspect of the present disclosure, and determining a resolution based on the predicted retention times and the identified peak widths for the one or more selected peptides; selecting one of the plurality of combinations of LC parameters having a highest resolution as a candidate combination of LC parameters; receiving a second input to specify a mobile phase gradient; adjusting the mobile phase gradient in the candidate combination of LC parameters to the specified mobile phase gradient, and reconstructing a UV chromatogram from the first preliminary UV chromatography data based on the adjusted candidate combination of LC parameters; and displaying a user interface containing the reconstructed UV chromatogram via the display.
[0011] According to a fifth aspect of the present disclosure, a computer system having a display is provided, which includes means for performing steps of the method according to the fourth aspect of the present disclosure.
[0012] According to a sixth aspect of the present disclosure, an electronic device is provided. The electronic device includes one or more processors, and a memory coupled to the one or more processors and storing computer-executable instructions which, when executed by the one or more processors, cause the one or more processors to perform the method according to any of the first to fourth aspects of the present disclosure.
[0013] According to a seventh aspect of the present disclosure, a non-transitory storage medium storing computer-executable instructions thereon is provided. The computer-executable instructions, when executed by one or more processors, cause the one or more processors to perform the method according to any of the first to fourth aspects of the present disclosure.
[0014] According to an eighth aspect of the present disclosure, a computer program product including instructions is provided. The instructions, when executed by a processor, implement the method according to any of the first to fourth aspects of the present disclosure.
[0015] According to a ninth aspect of the present disclosure, a computer program is provided. The computer program causes a computer to implement the method according to any of the first to fourth aspects of the present disclosure.BRIEF DESCRIPTION OF THE DRAWINGS
[0016] From the following description of embodiments of the present disclosure shown in conjunction with the drawings, the foregoing and other features and advantages of the present disclosure will become clear. The drawings are incorporated herein and form a part of the specification, which are further used to explain the principles of the present disclosure and enable those skilled in the art to make and use the present disclosure.
[0017] Fig. 1 is a flowchart showing a method for predicting a retention time (RT) for a peptide according to some embodiments of the present disclosure.
[0018] Fig. 2 is a schematic block diagram showing a neural network model for predicting a retention time (hereinafter briefly referred to as the RT model) that can be used in the method for predicting a retention time for a peptide according to some embodiments of the present disclosure.
[0019] Fig. 3 is a schematic block diagram showing a non-limiting example implementation of an encoder included in the neural network model of Fig. 2.
[0020] Fig. 4 is a schematic block diagram showing a prediction head included in the neural network model of Fig. 2 according to some embodiments of the present disclosure.
[0021] Fig. 5 is a schematic block diagram showing a non-limiting example implementation of a parameter prediction network included in the prediction head of Fig. 4.
[0022] Fig. 6 is a schematic block diagram showing a non-limiting example implementation of a parameter prediction network included in the prediction head of Fig. 4.
[0023] Fig. 7 is a schematic block diagram showing a non-limiting example implementation of a parameter prediction sub-network included in the parameter prediction network of Fig. 6.
[0024] Fig. 8 is a schematic block diagram showing a prediction head included in the neural network model of Fig. 2 according to some embodiments of the present disclosure.
[0025] Fig. 9 exemplarily illustrates a design space for LC parameters with each point representing a combination of LC parameters (herein also referred to as the LC condition) .
[0026] Fig. 10 shows performances of two RT models, one equipped with the prediction head of Fig. 4 and the other equipped with the prediction head of Fig 8, with training and validation loss curves at the top, plots of predicted RTs versus Measured RTs at the middle, and error distributions at the bottom.
[0027] Fig. 11 is a flowchart showing a method for predicting intrinsic parameters (IP) of a retention theory equation for a peptide according to some embodiments of the present disclosure.
[0028] Fig. 12 is a schematic block diagram showing a neural network model for predicting intrinsic parameters of a retention theory equation (hereinafter briefly referred to as the IP model) that can be used in the method for predicting intrinsic parameters of a retention theory equation for a peptide according to some embodiments of the present disclosure.
[0029] Fig. 13 is a flowchart showing a method for predicting a retention time for a peptide according to some embodiments of the present disclosure.
[0030] Fig. 14 is a flowchart showing a method according to some embodiments of the present disclosure.
[0031] Fig. 15 shows an example user interface (UI) for optimizing LC parameters in peptide mapping.
[0032] FIG. 16 shows a predicted UV chromatogram from fitting at the top and a measured chromatogram at the bottom, with an asterisk (*) indicating a solvent peak and a hash (#) marking a low-abundance N-glycan modification.
[0033] Fig. 17 shows UV chromatograms before and after the optimization.
[0034] Fig. 18 illustrates a non-limiting example two-phase process of optimizing LC parameters for peptide separation.
[0035] Fig. 19 is a schematic block diagram showing an electronic device according to some embodiments of the present disclosure.
[0036] Fig. 20 is a schematic block diagram showing a computer system, upon which embodiments of the present disclosure may be implemented.
[0037] Note that in the implementations illustrated below, sometimes the same reference numerals are commonly used across different drawings to denote the same parts or parts with the same function, and repeated descriptions thereof are omitted. In some cases, similar numbers and letters are used to denote similar items. Therefore, once an item is defined in a drawing, it does not need to be further discussed in subsequent drawings.
[0038] For ease of understanding, positions, dimensions, ranges, and the like of the structures shown in the drawings or the like sometimes do not represent actual positions, dimensions, ranges, and the like. Therefore, the present disclosure is not limited to the positions, dimensions, ranges, and the like disclosed in the drawings or the like.DETAILED DESCRIPTION
[0039] Various exemplary embodiments of the present disclosure will be described below in detail with reference to the drawings. It should be noted that: unless otherwise specifically illustrated, numerical expressions, numerical values, and relative arrangements of components and steps set forth in these embodiments do not limit the scope of the present disclosure.
[0040] In fact, the following description of at least one exemplary embodiment is merely illustrative, and in no way constitutes any limitation on the present disclosure and the application or use thereof. In other words, structures and methods herein are shown in an exemplary manner to illustrate different embodiments of the structures and the methods in the present disclosure. However, those skilled in the art will understand that they merely illustrate exemplary manners to implement the present disclosure rather than exhaustive ones. Moreover, the drawings are not necessarily drawn to scale, and some features may be enlarged to show details of specific components.
[0041] In addition, technologies, methods, and devices known to a person of ordinary skill in the related art may not be discussed in detail, but in appropriate cases, the technologies, methods, and devices shall be regarded as a part of the specification.
[0042] In all examples that are shown and discussed herein, any specific value should be interpreted only as an example but not as a limitation. Therefore, there may be different values for other examples of the exemplary embodiments.
[0043] It’s important to optimize LC parameters (including various variables in LC methods) in order to achieve effective peptide separation in LC methods for peptide mapping. Conventionally, a one-factor-at-a-time (OFAT) approach has been commonly used to optimize LC parameters, but the OFAT approach often overlooks potential interactions and synergistic effects among the variables. As a result, the OFAT approach can be a labor-intensive, time-consuming, and iterative process. In recent years, more comprehensive and systematic approaches based on Design of Experiments (DoE) have gained increasing attention in the field of peptide mapping. DoE approaches address the complex interplay of several variables by underlying statistical principles and retention models, allowing peptide map prediction in the design space. There are several well-known commercial platforms that utilize DoE for optimization and analysis, including Fusion and These platforms provide advanced tools and algorithms to facilitate the implementation of DoE in the development and optimization of LC methods. However, it is important to note that the DoE approaches still require a significant number of experiments, typically around a dozen, to explore the design space. This reliance on preliminary data still means high cost in terms of labor and time.
[0044] Retention time (RT) is a key factor in LC. The retention time of a peptide refers to the time point it eludes from a LC column. Predicting the retention time for the peptide can help overcome the limitations of the above approaches that rely heavily on experimental data. The present disclosure provides a method for predicting a retention time of a peptide that takes various LC parameters into account, upon which a method for optimizing LC parameters for peptide separation is further provided.
[0045] Various embodiments of the method for predicting a retention time for a peptide according to the present disclosure are described below in detail with reference to the accompanying drawings. It should be understood that the actual method may further include some other steps, however, these other steps are neither discussed herein nor illustrated in the drawings in order to avoid obscuring the key points of the present disclosure.
[0046] Fig. 1 is a flowchart showing a method 100 for predicting a retention time for a peptide according to some embodiments of the present disclosure. As shown in Fig. 1, the method 100 includes steps S102 to S106.
[0047] At S102, an embedding of an amino acid sequence of a peptide is obtained.
[0048] Since the amino acid sequence of the peptide is not numerical, it need to be vectorized so as to generate its embedding. Specifically, the amino acid sequence of the peptide has components such as amino acids and modifications, each of which may be assigned with a unique numerical value. For example, each of the twenty standard amino acids, represented by the characters “ACDEFGHIKLMNPQRSTVWY” respectively, may be assigned an integer value ranging from 1 to 20. The modifications may be treated similarly. In some examples, only a common post-translational modification, which is the cyclization of the N-terminal glutamine into pyroglutamic acid, is considered in addition to the standard amino acids, while other modifications that are less prevalent are excluded due to their minimal UV response in the UV chromatograms. For example, the cyclization of the N-terminal glutamine into pyroglutamic acid may be assigned an integer value of 21. As such, the amino acid sequence of the peptide can be converted into a one-dimensional array, or say a vector. For example, consider a peptide having an amino acid sequence of "VYACEVTHQGLSSPVTK" may be vectorized as [18, 20, 1, 2, 4, 18, 17, 7, 14, 6, 10, 16, 16, 13, 18, 17, 9] .
[0049] It’s also possible to use one-hot encoding to encode each component of the amino acid sequence into a vector, and thus the amino acid sequence of the peptide can be converted into a two-dimensional array, or say a matrix. It should be understood that these vectorization approaches are illustrative but not restrictive, and other suitable vectorization approaches are also feasible herein.
[0050] Any suitable embedding techniques may be applied here. In some embodiments, an initial embedding of the amino acid sequence of the peptide is obtained by inputting the vectorized amino acid sequence into embedding layers. In the above example, each integer representing an amino acid or a modification is transformed by the embedding layers into a vector of an appropriate length for an encoder to be described later. Position indexes of the amino acids or modifications may be embedded in a similar manner, so as to obtain a positional embedding, which is then combined with (e.g., added to) the initial embedding to create the complete embedding of the amino acid sequence of the peptide. For example, the longest amino acid sequence may have a length of 70, meaning that there are 70 position indexes in all, each being assigned an integer value ranging from 1 to 70. As such, position indexes of the peptide can be vectorized and then input to the embedding layers to generate the positional embedding. It can be understood that for ease of processing, both the vectorized amino acid sequence and the vectorized position indexes may be padded with 0 to achieve a length of 70 before being input to the embedding layers.
[0051] At S104, LC parameters to be used for separation of the peptide are obtained.
[0052] The LC parameters may include any suitable variables in LC methods, such as the type of column, mobile phase composition, pH, additives, column temperature, mobile phase gradient (which may be sometimes briefly referred to herein as gradient) , and flow rate. For illustrative purpose, the temperature, mobile phase gradient, and flow rate will be taken as examples of the LC parameters hereinafter, since they are the most user-friendly variables that can be changed on-line.
[0053] For example, when the peptide "VYACEVTHQGLSSPVTK" is to be subjected to a separation under a LC condition with a mobile phase gradient having a duration of 180 minutes, a column temperature of 80℃, and a flow rate of 0.2 mL / min, a vector for the LC parameters to be input into a model to be described later may be delineated as [180.0, 80.0, 0.2] , with these floating-point numbers representing the mobile phase gradient duration, column temperature, and flow rate, respectively.
[0054] At S106, the embedding and the LC parameters are input to a neural network model (the RT model) trained for predicting a retention time, so as to obtain the retention time for the peptide. For example, in the above example for the peptide "VYACEVTHQGLSSPVTK" under the LC condition [180.0, 80.0, 0.2] , the RT model may output [71.4] , representing a retention time of 71.4 minutes.
[0055] Fig. 2 shows a neural network model for predicting a retention time (the RT model 200) that can be used in S106. As shown in Fig. 2, the RT model 200 includes an encoder 210 and a prediction head 220. The encoder 210 is configured to encode the embedding (obtained at S102) to generate an encoded embedding. The prediction head 220 is configured to generate a retention time for the peptide based on the encoded embedding and the LC parameters (obtained at S104) .
[0056] The encoder 210 can be implemented using various types of neural networks, including but not limited to Convolutional Neural Networks (CNNs) , Long Short-Term Memory (LSTM) networks, Gated Recurrent Units (GRUs) , and Transformers. Fig. 3 shows a non-limiting example implementation of the encoder 210, which is based on the Transformer architecture. The encoder 210 includes a number (e.g., N) of encoder blocks 2100 stacked serially each having the illustrated structure. Specifically, each encoder block 2100 has a first normalization module 2102, a multi-head attention module 2104, a first dropout module 2106, a second normalization module 2108, a feed-forward network module 2110, and a second dropout module 2112 arranged in series. There are a first residual connection 2114 from an input of the first normalization module 2102 to an output of the first dropout module 2106 and a second residual connection 2116 from an input of the second normalization module 2108 to an output of the second dropout module 2112. Alternatively, the first normalization module 2102 and the multi-head attention module 2104 may swap their positions, and the second normalization module 2108 and the feed-forward network module 2110 may also swap their positions. Dropout is a regularization technique where a certain percentage of neurons are randomly "dropped out" (set to zero) during training to prevent overfitting. In some cases, the dropout modules may be removed.
[0057] The hyperparameters used by the encoder 210 may be chosen according to actual needs. In some examples, the number N of the encoder blocks ranges from 8 to 12, e.g., N=10. In some examples, a number of attentions heads ranges from 6 to 10, e.g., the number being 8. In some examples, a dimension of the feed-forward network ranges from 512 to 2048, e.g., the dimension being 1024. In some examples, a dropout rate ranges from 0.05 to 0.3, or from 0.05 to 0.15, e.g., the dropout rate being 0.1.
[0058] Fig. 4 shows a non-limiting example implementation 220A of the prediction head 220 that can be used in the RT model 200. The prediction head 220A includes a parameter prediction network 222 and a retention time calculation unit 224.
[0059] The parameter prediction network 222 is configured to generate intrinsic parameters of a retention theory equation (which may also be referred to as retention-related intrinsic parameters) for the peptide based on the encoded embedding. The retention theory equation defines a retention time as a function of the LC parameters and the intrinsic parameters, e.g., RT = Function (LC parameters; intrinsic parameters) . Specifically, the parameter prediction network 222 derives the intrinsic parameters fundamental to the retention theory from the encoded embedding via neural networks. These intrinsic parameters effectively characterize the variations in RT as influenced by different LC parameters via the retention theory equation, which will be described in detail later.
[0060] The retention time calculation unit 224 is configured to calculate a retention time for the peptide based on the predicted intrinsic parameters and the obtained LC parameters through the retention theory equation. Consequently, the prediction head 220A integrates the mechanisms of LC separation into the neural networks, achieving a remarkable accuracy in predicting RTs under various LC conditions with different LC parameters.
[0061] Specifically, the encoded embedding is transformed into the intrinsic parameters through the parameter prediction network 222. In some embodiments, the parameter prediction network 222 includes a limit scale module configured to limit an output of the parameter prediction network 222 to working ranges of the intrinsic parameters of the retention theory equation. In other words, the limit scale module may scale its received input such that values of elements included in the final output fall within the working ranges of the respective intrinsic parameters. This helps to precisely derive the intrinsic parameters from the encoded embedding. In some examples, the limit scale module may store a set of limit scale coefficients (e.g., in a form of an array) , each of which corresponds to a respective one of the intrinsic parameters. The limit scale module may then scale its received input by applying the set of limit scale coefficients thereto (e.g., via multiplication) . For example, these limit scale coefficients may be determined through fitting of experimental data.
[0062] In some embodiments, the parameter prediction network 222 includes a plurality of linear activation modules located upstream of the limit scale module. Each linear activation module may have a linear layer combined with an activation function. In some examples, the last one of the plurality of linear activation modules may have a different activation function from others of the plurality of linear activation modules.
[0063] Moreover, in some embodiments, every adjacent two of the plurality of linear activation modules may have a dropout module arranged therebetween to enhance robustness and prevent overfitting. Each dropout module may have a dropout layer.
[0064] Fig. 5 shows a non-limiting example implementation 222A of the parameter prediction network 222 that can be used in the prediction head 220A. As shown in Fig. 5, the parameter prediction network 222A is designed with three linear activation modules 2224 interspersed with dropout modules 2226. The first two linear activation modules 2224 each have a linear layer equipped with a Gaussian Error Linear Unit (GELU) activation function. The last one linear activation module has a linear layer equipped with a Sigmoid activation function, and is followed by a limit scale module 2222. An output of the last one linear activation module is thus scaled by a predefined limit to exactly correspond to the working ranges of the intrinsic parameters of the retention theory equation. Specifically, an output of the Sigmoid activation function ranges from 0 to 1, and then the limit scale module 2222 receives and multiplies the output of the Sigmoid activation function with a predetermined set of limit scale coefficients so that a value of each element of an output of the limit scale module 2222 falls within a reasonable range for a respective one of the intrinsic parameters.
[0065] The retention theory equation typically includes a plurality of intrinsic parameters. In some embodiments, the parameter prediction network 222 includes a corresponding plurality of parameter prediction sub-networks arranged in parallel, each of which configured to generate a respective one of the plurality of intrinsic parameters based on the encoded embedding.
[0066] Fig. 6 shows another non-limiting example implementation 222B of the parameter prediction network 222 that can be used in the prediction head 220A. As shown in Fig. 6, the parameter prediction network 222B includes parameter prediction sub-networks 1, …, n arranged in parallel, with n being the number of the intrinsic parameters of the retention theory equation. Each of the parameter prediction sub-networks 1, …, n receives the encoded embedding and outputs a respective intrinsic parameter, which is then received by the retention time calculation unit 224. The retention time calculation unit 224 uses the received intrinsic parameter 1, …, n along with the obtained LC parameters to derive a retention time based on the retention theory equation.
[0067] Each of the parameter prediction sub-networks 1, …, n may have a structure similar to that of the parameter prediction network 222A. Specifically, Fig. 7 shows a non-limiting example implementation of a parameter prediction sub-network i (i=1, …, n) . The parameter prediction sub-network i may include three linear activation modules interspersed with dropout modules, and further include a limit scale module to scale an output of the parameter prediction sub-network i to a working range that corresponds to an intrinsic parameter i, so that the intrinsic parameter i may be derived reasonably.
[0068] The retention theory equation can be established according to any chromatographic principles, either known at present or developed in future, and the intrinsic parameters can be those in the retention theory equation that are not LC parameters, so that they can be integrated with the LC parameters to derive a retention time from the retention theory equation.
[0069] In some embodiments, the retention theory equation is derived from a linear solvent strength (LSS) theory equation which is expressed as:
[0070]
[0071] wherein k is a retention factor, k0 is the retention factor when (in other words, it’s an extrapolated value of retention factor in the solvent of the weakest elution strength in the LC (for example, water in the RPLC) ) , S is a slope constant in the LSS theory equation (also known as the solvent strength parameter) , is a fraction of a component of a mobile phase in LC (usually expressed as volumetric fraction) .
[0072] In some embodiments, the intrinsic parameters of the retention theory equation comprise S and coefficients of an expression of k0 or ln k0 with respect to a LC parameter (s) of the retention theory equation. k0 or ln k0 can be expressed as any suitable function of the LC parameter (s) of the retention theory equation. In some examples, k0 or ln k0 can be expressed as an expansion (for example without limitation, a Taylor expansion) with respect to the LC parameter (s) of the retention theory equation. Expressing k0 or ln k0 with respect to the LC parameter (s) can introduce dependence on the LC parameter (s) into the equation. Such LC parameter (s) can be selected as needed. For example, the LC parameters of the retention theory equation include a mobile phase gradient indicating as a function of time t, a column temperature T, and a flow rate F, and the intrinsic parameters of the retention theory equation include S and coefficients of an expansion of k0 or ln k0 with respect to T.
[0073] Expanding k0 or ln k0 with respect to T can introduce temperature dependence into the equation. As a non-limiting example, a Taylor expansion of ln k0 is as follows:
[0074]
[0075] For gradient elution, a retention time tR of an analyte follows the below fundamental integral equation:
[0076]
[0077] wherein t0=V0 / F is the holdup time that the mobile phase passes through the column, V0 is the holdup volume which may be a known constant, and F is the flow rate.
[0078] For this particular example, the set of equations (1) - (3) leads to as the retention theory equation with three LC parameters T, F and four intrinsic parameters S, A, B, C. Using these four intrinsic parameters S, A, B, C derived from the encoded embedding for a specific peptide by the parameter prediction network 222 and the retention theory equation the retention time calculation unit 224 can predict a retention time for the specific peptide with varied mobile phase gradient column temperature T, and flow rate F.
[0079] In the above equation (2) , only first three items of the Taylor expansion are kept while the remaining items are omitted. It should be understood that a person of ordinary skills in the relevant art may choose which items to be omitted according to actual circumstances. In some cases, keeping more than three items may not be desired, as it might introduce too much covariance among parameters. In some cases, keeping less than three items may also not be desired, as the introduced temperature dependence might be insufficient.
[0080] It’s possible not to expand k0 or ln k0 with respect to T if there is no need to consider the column temperature. In this case, the set of equations (1) and (3) leads to as the retention theory equation with two LC parameters F and two intrinsic parameters S, k0.
[0081] Moreover, it’s also possible that the set of equations (1) - (3) leads to as the retention theory equation with three LC parameters T, F and five intrinsic parameters S, A, B, C, k0. However, this is not so recommended, as there are increased dependences of parameters (specifically, S, A, B, C, k0) on each other.
[0082] Alternatively, the encoded embedding may be directly merged with LC parameters within a neural network to predict the retention time. Specifically, Fig. 8 shows another non-limiting example implementation 220B of the prediction head 220 that can be used in the RT model 200. The prediction head 220B includes convolution layers 2262 configured to perform feature extraction on the encoded embedding, a flattened layer 2264 located downstream of the convolution layers 2262 and configured to convert an output of the convolution layers 2262 to a one-dimensional vector, and fully connected layers 2266 located downstream of the flattened layer 2264 and configured to receive an output of the flattened layer 2264 and the LC parameters and output a retention time for the peptide.
[0083] In some embodiments, the RT model 200 is trained by a process including the following steps: identifying an amino acid sequence and retention time actual values of a sample peptide separated with a plurality of different combinations of LC parameters; obtaining an embedding of the identified amino acid sequence of the sample peptide (e.g., in the same manner as S102) ; inputting the embedding of the identified amino acid sequence of the sample peptide and the LC parameters for separation of the sample peptide to the RT model 200, so as to obtain retention time predicted values for the sample peptide; and updating parameters of the RT model 200 through a loss function based on the retention time actual values and the retention time predicted values of the sample peptide. It should be understood that the loss function may adopt any suitable form that is known at present or developed in future, such as but not limited to a Mean Squared Error (MSE) loss function or a Mean Absolute Error (MAE) loss function.
[0084] Specifically, the sample peptide may be subjected to liquid chromatography-tandem mass spectrometry (LC-MS / MS) . The sample peptide may be separated under different LC conditions, each corresponding to a respective one of a plurality of different combinations of LC parameters. After separation, the sample peptide may be detected by a mass spectrometer operating with a full MS scan followed by MS / MS scans. The amino acid sequence and retention time actual values of the sample peptide may be identified from the MS data.
[0085] For the purpose of illustration, a non-limiting example process of preparing datasets for training the RT model 200 is described herein. In this example, stationary and mobile phases typically utilized for peptide mapping were selected. Here, a commercial column (Agilent / Poroshell 120 SB-C18, 2.1×150 mm, 2.7 μm) was chosen. Water and acetonitrile were two components A and B constituting the mobile phase, respectively, and thus would be a fraction, B%, of the component B that is acetonitrile. The addition of an ion pairing agent, such as trifluoroacetic acid (TFA) , can enhance the retention of peptides, especially those with a larger number of charges. Variations in the concentration of TFA can lead to shifts in RTs. To ensure the robustness of the mobile phase preparation, 1%TFA was deliberately selected. At this concentration, the ion pairing effect reaches near saturation. Mobile phase gradient, column temperature, and flow rate were selected as LC parameters to be varied.
[0086] A design space for the LC parameters was designed within the constraints dictated by the operational limits of instruments and the scope of practical applications. The mobile phase gradient parameter was defined with B%ascending from 0 to 40%over an interval of 120 to 180 minutes. The column temperature was confined within a calibrated thermal range of 50 to 80 degrees Celsius, while the flow rate was set to fluctuate between 0.2 and 0.4 mL / min. To construct an empirically robust training dataset, the design space was systematically sampled at 27 points, capturing the lower, median, and upper thresholds of each parameter, as shown in Fig. 9.
[0087] For preparing the datasets, RTs of peptides digested by Trypsin and LysC from whole Chinese Hamster Ovary (CHO) proteins were measured. The proteolytic peptides were separated by RPLC under 27 LC conditions represented by the 27 points in Fig. 9, respectively, and were further detected by a mass spectrometer operating with full MS scan followed by MS / MS scans. Peptides (amino acid sequences thereof) and RTs were identified using the Proteome Discoverer software, which employs the Sequest HT search engine (Thermo Fisher Scientific Inc. ) .
[0088] Table 1
[0089]
[0090] The datasets were randomly divided into training, validation, and test segments as delineated in Table 1. The training dataset was utilized to construct and enhance the predictive capabilities of the model by learning from a comprehensive set of known amino acid sequences of peptides and LC conditions. For the validation and test datasets, the amino acid sequences remained unexposed to the model to ensure unbiased evaluation. The validation dataset served a dual purpose: it was utilized to observe the loss progression and acted as a criterion for selecting the epoch that demonstrated optimal performance. The test dataset, on the other hand, was employed to assess predictive accuracy post-training. Additionally, one specific LC condition was deliberately withheld from the training phase to reserve it for testing predictive prowess of the model under previously unseen LC conditions.
[0091] Table 2
[0092]
[0093] The hyperparameters used for the encoder 210 and two prediction heads 220A, 220B mentioned above are listed in Table 2. The prediction performances of two RT models equipped with the encoder 210 and two prediction heads 220A, 220B, respectively, were evaluated and compared through the inclusion of loss curves, predicted versus measured plots, and error distributions as shown in Fig. 10. The loss curves chart the absolute errors over the course of training iterations, which indicates both models have achieved a state of robust convergence and stability. The predicted versus measured plots using the test dataset visually demonstrate the high accuracy with correlation coefficients exceeding 0.97. The close alignment along the line of perfect agreement (the 45-degree line) suggests a high level of precision, thereby validating the predictive power of the model. The error distribution analysis of the test dataset elucidates patterns in the residuals, facilitating a comparative assessment of the models. The error distribution of the reserved test dataset aligns closely with that of the primary test dataset, indicating that the model proficiently predicts RTs for novel peptides even under unexplored LC conditions.
[0094] In the prediction head 220A, the encoded embeddings initially predict the intrinsic parameters that describe the retention behaviors of peptides under different LC conditions. Subsequently, these intrinsic parameters are integrated with the retention theory equation to forecast the retention times for various combinations of LC parameters. The prediction 220A that uses encoded embeddings to predict intrinsic parameters and then integrates the intrinsic parameters with LC parameters via the retention theory to predict the retention time (Fig. 10 a) ) is superior to the prediction 220B that directly merges the encoded embedding with LC parameters within pure neural networks to predict the retention times (Fig. 10 b) ) . Compared with the prediction head 220B, the prediction 220A provides more nuanced understanding of the intrinsic parameters, leading to more precise predictions under variable LC conditions. Crucially, the prediction head 220A can provide a high level of adaptability that retention times from any mobile phase gradient shape can be forecasted accurately.
[0095] Accordingly, with reference to Fig. 11, the present disclosure further provides a method 300 for predicting intrinsic parameters of a retention theory equation for a peptide. As described above, the retention theory equation defines a retention time as a function of LC parameters and the intrinsic parameters. As shown in Fig. 11, the method 300 includes steps S302 and S304.
[0096] At S302, an embedding of an amino acid sequence of a peptide is obtained. S302 is similar to S102, and thus repetitive description thereof is omitted here.
[0097] At S304, the embedding is input to a neural network model (the IP model) trained for predicting intrinsic parameters of the retention theory equation, so as to obtain the intrinsic parameters for the peptide.
[0098] Fig. 12 shows a neural network model for predicting intrinsic parameters (the IP model 400) that can be used in S304. As shown in Fig. 12, the IP model 400 includes an encoder 410 and a parameter prediction network 420. The encoder 410 is configured to encode the embedding (obtained at S302) to generate an encoded embedding. The parameter prediction network 420 is configured to generate intrinsic parameters for the peptide based on the encoded embedding.
[0099] The encoder 410 is similar to the encoder 210, and thus repetitive description thereof is omitted here. Also, the parameter prediction network 420 is similar to the parameter prediction network 222, and thus repetitive description thereof is omitted here.
[0100] In some embodiments, the IP model 400 is trained by a process including the following steps: identifying an amino acid sequence and retention time actual values of a sample peptide separated with a plurality of different combinations of LC parameters; obtaining an embedding of the identified amino acid sequence of the sample peptide (e.g., in the same manner as S302) ; inputting the embedding of the identified amino acid sequence of the sample peptide to the IP model 400, so as to obtain the intrinsic parameters for the sample peptide; determining retention time predicted values of the sample peptide based on the obtained intrinsic parameters for the sample peptide and the LC parameters for separation of the sample peptide through the retention theory equation; and updating parameters of the IP model 400 through a loss function based on the retention time actual values and the retention time predicted values of the sample peptide. It should be understood that the loss function may adopt any suitable form that is known at present or developed in future, such as but not limited to a MSE loss function or a MAE loss function.
[0101] Specifically, the sample peptide may be subjected to LC-MS / MS. The sample peptide may be separated under different LC conditions, each corresponding to a respective one of a plurality of different combinations of LC parameters. After separation, the sample peptide may be detected by a mass spectrometer operating with a full MS scan followed by MS / MS scans. The amino acid sequence and retention time actual values of the sample peptide may be identified from the MS data. A process for preparing datasets for training the IP model 400 may similarly refer to that for the RT model 200, and thus repetitive description thereof is omitted here.
[0102] A method for predicting a retention time for a peptide is provided based on the method 300. Specifically, after intrinsic parameters for the peptide are obtained using the method 300 and LC parameters to be used for separation of the peptide are obtained, a retention time for the peptide is calculated based on the obtained intrinsic parameters and the obtained LC parameters through the retention theory equation. For example, Fig. 13 shows a method 500 for predicting a retention time for a peptide, which includes steps S502 to S508.
[0103] At S502, an embedding of an amino acid sequence of a peptide is obtained. S502 is similar to S102 and S302, and thus repetitive description thereof is omitted here.
[0104] At S504, the embedding is input to a neural network model (e.g., the IP model 400) trained for predicting intrinsic parameters of a retention theory equation that defines a retention time as a function of LC parameters and the intrinsic parameters, so as to obtain the intrinsic parameters for the peptide. S504 is similar to S304, and thus repetitive description thereof is omitted here.
[0105] At S506, LC parameters to be used for separation of the peptide are obtained. S506 is similar to S104, and thus repetitive description thereof is omitted here.
[0106] At S508, a retention time for the peptide is calculated based on the obtained intrinsic parameters and the obtained LC parameters through the retention theory equation. Details on the retention theory equation may refer to the above description and thus repetitive description thereof is omitted here.
[0107] With the methods 100, 300, 500 and / or the models 200, 400, the present disclosure further provides a method to utilize this capability of RT prediction to optimize LC parameters for peptide separation as well as to refine UV chromatograms with a low burden on the preparation of preliminary data. A predictive platform that is built based on the models 200 and / or 400 is also provided, with a user-friendly interface for optimization of LC parameters and visual examination of real-time predicted UV chromatograms and peptide information tables including details on sequences, retention times, and resolutions.
[0108] Fig. 14 shows a method 600 that may be performed at a computer system having a display. Accordingly, the present disclosure also provides a computer system having a display which includes means for performing steps of the method 600. The method 600 includes steps S602 to S614.
[0109] At S602, a first input containing first preliminary UV chromatography data and first MS data of a plurality of peptides separated with a first combination of LC parameters is received. Peak areas and widths and amino acid sequences of the plurality of peptides are identified from the first preliminary UV chromatography data and the first MS data.
[0110] Various computational tool or software that assists in the identification of peptides based on their MS data may be utilized, such as Protein Metrics Inc (PMI) software which can identify peptides by comparing the measured mass-to-charge ratios (m / z) from MS data against known peptide databases.
[0111] For example, the plurality of peptides may be separated by RPLC with the first combination of LC parameters, and their preliminary UV chromatography data may be acquired with a detector set at a wavelength of 214 nm. After separation, these peptides may be further detected by a mass spectrometer operating with full MS scan followed by MS / MS scan, thereby generating their MS data.
[0112] In some embodiments, an input containing second preliminary UV chromatography data of the plurality of peptides separated with a second combination of LC parameters that is different from the first combination of LC parameters is received. The peak areas and widths of the plurality of peptides may be identified from the first and second preliminary UV chromatography data.
[0113] Specifically, if all peaks are resolved in the first preliminary UV chromatography data, then only the first preliminary UV chromatography data would be sufficient to identify peak areas and widths for all the peptides. When at least some of the peaks are unresolved in the first preliminary UV chromatography data, the second preliminary UV chromatography data can help the first preliminary UV chromatography data to manage those unresolved peaks. For example, for peptides with peaks that are either resolved or unresolved with a resolution cutoff of 1.2 in both the first and second preliminary UV chromatography data, average peak areas and widths on the first and second preliminary UV chromatography data are assigned to these peptides. For peptides resolved exclusively in one of the first and second preliminary UV chromatography data, peak areas and widths identified from the one of the first and second preliminary UV chromatography data are assigned to these peptides. It’s understood that more preliminary UV chromatography data may be similarly involved to manage unresolved peaks according to actual needs. Typically, two trials would be sufficient.
[0114] At S604, a plurality of combinations of LC parameters is generated. Each of the plurality of combinations of LC parameters may have a default mobile phase gradient, which may be the mobile phase gradient of the first combination of LC parameters or any other suitable mobile phase gradient. In other words, all of the plurality of combinations of LC parameters have the same mobile phase gradient. Other LC parameters than the mobile phase gradient in each of the plurality of combinations of LC parameters may correspond to a respective point in a parameter design space with each dimension corresponding to a respective one of the other LC parameters. For example, assuming that the combinations of LC parameters consist of mobile phase gradient, temperature, and flow rate, then a two-dimensional parameter design space with one dimension being temperature and the other dimension being flow rate can be constructed within the constraints dictated by the operational limits of instruments and the scope of practical applications.
[0115] At S606, for each of the plurality of combinations of LC parameters, retention times for one or more selected peptides of the plurality of peptides are predicted using the method 100 or the method 500, and a resolution is determined based on the predicted retention times and the identified peak widths for the one or more selected peptides.
[0116] In some embodiments, the resolution is determined for each of the plurality of combinations of LC parameters by: for each of the one or more selected peptides, determining a resolution for the peptide based on the predicted retention time and the identified peak width for the peptide; and determining a lowest resolution among the resolutions for the one or more selected peptides as the resolution for the combination of LC parameters.
[0117] For example, the resolution Rs for the peptide may be defined as:
[0118]
[0119] Wherein RT1 and are RT2 retentions times of the peptide and its nearest peptide in the UV chromatogram, respectively, and FWHM1 and FWHM2 are full widths at half maximum (FWHMs) of peaks of the peptide and its nearest peptide in the UV chromatogram, respectively.
[0120] As such, a resolution map may be generated with the same dimensions as the parameter design space, which can visualize the variation of resolutions with the LC parameters.
[0121] At S608, one of the plurality of combinations of LC parameters having a highest resolution is selected as a candidate combination of LC parameters. At this point, the other parameters than the mobile phase gradient in the candidate combination of LC parameters may be considered to have been optimized. In other words, the candidate combination of LC parameters now has the optimized temperature, the optimized flow rate, and the default mobile phase gradient, for example.
[0122] At S610, a second input to specify a mobile phase gradient is received. For example, this second input may be provided from a user.
[0123] At S612, the mobile phase gradient in the candidate combination of LC parameters is adjusted to the specified mobile phase gradient, and a UV chromatogram is reconstructed from the first preliminary UV chromatography data based on the adjusted candidate combination of LC parameters.
[0124] In some embodiments, the UV chromatogram is reconstructed from the first preliminary UV chromatography data based on the adjusted candidate combination of LC parameters by: for the adjusted candidate combination of LC parameters, predicting retention times for the plurality of peptides using the method 100 or method 500; and generating the reconstructed UV chromatogram based on the identified peak areas and widths and the predicted retention times of the plurality of peptides. As the adjustment to the mobile phase gradient could influence the RTs which determine UV peak positions, it would stretch or compress the UV peaks within specific local regions.
[0125] At S614, a user interface containing the reconstructed UV chromatogram is displayed via the display. As such, alterations in the UV chromatogram in response to adjustments made to the LC parameters, especially the mobile phase gradient, can be directly visualized to the user, thereby providing immediate feedback on the influence of LC parameter modifications and enhancing the precision of analytical outcomes.
[0126] In some embodiments, a third input to save or otherwise confirm the reconstructed UV chromatogram displayed on the user interface is received. In this case, a combination of LC parameters corresponding to the reconstructed UV chromatogram that is saved or otherwise confirmed is determined as LC parameters recommended for separation of the plurality of peptides. At this point, the mobile phase gradient may be considered to have been optimized, which combines with the other LC parameters that have been optimized in previous steps to constitute a recommended LC condition for separation of the peptides.
[0127] For illustrative purposes, Fig. 15 shows a non-limiting example user interface (UI) 700 in which the method of the present disclosure is applied. The UI 700 includes a input region 702, a peptide information region 704, a UV chromatogram region 706, a LC parameters region 708, and a resolution map region 710. Trypsin-digested commercial monoclonal antibody Tremfya is illustrated in Fig. 15 as an example, and stationary and mobile phases were chosen to be the same as the above example. Export from the PMI software and UV raw data were initially provided alongside the amino acid sequences of peptides (briefly referred to as the peptide sequences) in the Fasta filetype. Upon activating the “INIT” button (initialization) , the peptide information table displays the peptide sequences, RTs, and resolutions under a default initialization LC condition. Export from the PMI software contains the RTs and FWHMs for the peptides, which serve as an initial guess during a process of fitting the UV peak areas and widths to a Gaussian profile. As shown in Fig. 16, the predicted UV chromatogram resulting from the fitting corresponds well to the measured UV chromatogram. The predicted UV chromatogram was displayed in the UV chromatogram region 706 as a preliminary UV chromatogram.
[0128] Subsequently, unique peptides requiring enhanced separation may be selected. In this instance, peptides containing complementarity determining regions identified by labels 2, 6, 24, 32, and 37 were selected. Following this, the “OPTI” button (optimization) was clicked to generate the resolution map under a default linear mobile phase gradient, which was displayed in the resolution map region 710. This resolution map reflects the minimal resolutions of peaks for the selected peptides within the design space of the LC parameters targeted for optimization, specifically temperature and flow rate in this instance. In this scenario, the optimized temperature is 61 degrees Celsius, and the optimized flow rate is 0.2 mL / min. After this optimization, the RTs and resolutions in the peptide information region 704 were updated to predicted RTs under the optimized temperature, the optimized flow rate, and the default linear mobile phase gradient (e.g., using the method 100 or 500) and resolutions calculated based on the predicted RTs and the peak widths. The UV chromatogram displayed in the UV chromatogram region 706 highlights the peaks for the selected peptides with marks, and was updated to a reconstructed UV chromatogram based on the predicted RTs under the optimized temperature, the optimized flow rate, and the default linear mobile phase gradient from the previous UV chromatogram.
[0129] Then, segmented linear mobile phase gradients were adjusted interactively ( “ADJU” button and mobile phase gradient table in the LC Parameters region 708) , accompanied by real-time reconstructed UV chromatograms and real-time updated RTs and resolutions in the peptide information table. In this case, the finalized mobile phase gradient is documented in Table 3.
[0130] Table 3
[0131]
[0132] Ultimately, the peptide information table and the refined UV chromatogram can be downloaded by clicking the "SAVE" button. The "CLEA" button is utilized to reset the system, ensuring a fresh start for subsequent analyses. The refined UV chromatogram can be used for subsequent analysis.
[0133] The UV chromatograms before and after optimization are depicted in Fig. 17. The resolutions of the selected peaks are more distinctly separated after optimization, and the overall peak distribution has been enhanced with a reduced mobile phase gradient duration. This example underscores the successful application of this predictive platform for developing liquid chromatographic methods in peptide mapping, demonstrating its efficacy in refining analytical precision and operational efficiency in peptide analysis.
[0134] Fig. 18 summarizes a two-phase approach for optimizing LC parameters that is utilized in the above workflow. The first phase may be either phase 1 or phase 1’ , and the second phase may be phase 2. In phase 1, amino acid sequences, or more specifically their embeddings, are input to the RT model 200 disposed on the predictive platform, along with various combinations of LC parameters obtained from the parameter design space. Retention times from the RT model 200 are integrated with peak widths to generate a resolution map, from which optimal LC parameters other than the mobile phase gradient can be found.
[0135] Alternatively, in phase 1’ , amino acid sequences, or more specifically their embeddings, are input to the IP model 400 disposed on the predictive platform to derive intrinsic parameters associated with their retention behaviors. These intrinsic parameters are integrated with various combinations of LC parameters obtained from the parameter design space to generate corresponding retention times via the retention theory equation. The resulted retention times are integrated with peak widths to generate a resolution map, from which optimal LC parameters other than the mobile phase gradient can be found.
[0136] In phase 2, the mobile phase gradient is iteratively adjusted and real-time reconstructed UV chromatogram is accordingly displayed until a satisfied separation is realized. Specifically. the user may visually examine the UV chromatograms and decide whether to further adjust the mobile phase gradient or accept the mobile phase gradient.
[0137] This two-phase approach for optimizing LC parameters consumes less resources and time, and can be especially advantageous in enhancing efficiency and precision in peptide mapping, and streamlining the analytical workflow in biotherapeutic protein development. Compared to traditional OFAT optimization methods, this approach eliminates the guesswork and repetitive trial-and-error typically involved in LC separation for peptide mapping. Traditional methods often struggle due to complex interactions between LC parameters and diverse peptide solutions, resulting in inefficiency. This approach systematically analyzes and predicts the effects of varying parameter combinations, greatly enhancing the accuracy and efficiency of optimization, and providing a more scientific and effective strategy for LC separation. This approach circumvents the need for multiple preliminary experiments typically required by commercial DoE software during optimization. Preliminary experiments, which require substantial manual labor and significant instrument time, are further complicated by the need to track component peak RTs under varied LC conditions. Utilizing an advanced predictive model (e.g., the RT model 200 or the IP model 400) , this approach directly forecasts RTs across different LC conditions, drastically reducing experimental volume and boosting efficiency. This approach conserves resources while enhancing the accuracy and reliability of LC analysis, offering robust support for analyzing complex samples.
[0138] The present disclosure also provides an electronic device, which may include one or more processors and a memory coupled to the one or more processors and storing computer-executable instructions which, when executed by the one or more processors, cause the one or more processors to perform the method according to any aforementioned embodiment of the present disclosure. As shown in FIG. 19, an electronic device 900 includes a processor (s) 902 and a memory 904 storing computer-executable instructions which, when executed by the processor (s) 902, cause the processor (s) 902 to perform the method according to any aforementioned embodiment of the present disclosure. The processor (s) 902 may, for example, be a central processing unit (CPU) or graphics processing unit (GPU) of the electronic device 900. The processor (s) 902 may be any type of general-purpose processor, or may be a processor specially designed to predict a retention time or intrinsic parameters for a peptide, or to optimize LC parameters for peptide separation, such as an application specific integrated circuit ( “ASIC” ) . The memory 904 may be coupled to the processor (s) 902, and may include various computer-readable media that can be accessed by the processor (s) 902. In various embodiments, the memory 904 described herein may include volatile and non-volatile media as well as removable and non-removable media. For example, the memory 904 may include any combination of the following: a random access memory ( “RAM” ) , a dynamic RAM ( “DRAM” ) , a static RAM ( “SRAM” ) , a read-only memory ( “ROM” ) , a flash memory, a cache memory, and / or any other type of non-transitory computer-readable medium. The memory 904 may store instructions which, when executed by the processor 902, cause the processor 902 to perform the method according to any aforementioned embodiment of the present disclosure.
[0139] The present disclosure further provides a non-transitory storage medium storing computer-executable instructions thereon which, when executed by one or more processors, cause the one or more processors to perform the method according to any aforementioned embodiments of the present disclosure.
[0140] The present disclosure further provides a computer program product that may include instructions which, when executed by a processor, may implement the method according to any aforementioned embodiment of the present disclosure. The instructions may be any instruction set to be executed directly by one or more processors, such as machine codes, or any instruction set to be executed indirectly, such as scripts. The instructions can be stored in a format of object codes for direct processing by one or more processors, or stored in any other computer language, including scripts or collections of independent source code modules that are interpreted on demand or compiled in advance.
[0141] The present disclosure further provides a computer program causing a computer to implement the method according to any aforementioned embodiment of the present disclosure.
[0142] Fig. 20 is a schematic block diagram that illustrates a computer system 1000, upon which embodiments of the present disclosure may be implemented. The computer system 1000 includes a bus 1002 or other communication mechanism for communicating information, and a processor 1004 coupled with the bus 1002 for processing information. The computer system 1000 also includes a memory 1006, which can be a random access memory (RAM) or other dynamic storage device, coupled to the bus 1002 for storing instructions to be executed by the processor 1004. The memory 1006 also may be used for storing temporary variables or other intermediate information during execution of instructions to be executed by the processor 1004. The computer system 1000 further includes a read only memory (ROM) 1008 or other static storage device coupled to the bus 1002 for storing static information and instructions for the processor 1004. A storage device 1010, such as a magnetic disk or optical disk, is provided and coupled to the bus 1002 for storing information and instructions. The computer system 1000 may be coupled via the bus 1002 to a display 1012, such as a cathode ray tube (CRT) or liquid crystal display (LCD) , for displaying information to a computer user. An input device 1014, such as a keyboard, a mouse, or a touch panel, is coupled to the bus 1002 for communicating information and command selections to the processor 1004.
[0143] The computer system 1000 can perform the embodiments of the present disclosure. Consistent with certain implementations of the present disclosure, results are provided by the computer system 1000 in response to the processor 1004 executing one or more sequences of one or more instructions contained in the memory 1006. Such instructions may be read into the memory 1006 from another computer-readable medium, such as the storage device 1010. Execution of the sequences of instructions contained in the memory 1006 causes the processor 1004 to perform the process described herein. Alternatively, hard-wired circuitry may be used in place of or in combination with software instructions to implement the present disclosure. Thus implementations of the present disclosure are not limited to any specific combination of hardware circuitry and software.
[0144] In various embodiments, the computer system 1000 can be connected via a network interface 1016 to one or more other computer systems, like the computer system 1000, across a network to form a networked system. The network can include a private network or a public network such as the Internet. In the networked system, one or more computer systems can store and serve the data to other computer systems. The one or more computer systems that store and serve the data can be referred to as servers or the cloud, in a cloud computing scenario. The one or more computer systems can include one or more web servers, for example. The other computer systems that send and receive data to and from the servers or the cloud can be referred to as client or cloud devices, for example.
[0145] The term “computer-readable medium” as used herein refers to any media that participates in providing instructions to the processor 1004 for execution. Such a medium may take many forms, including but not limited to, non-volatile media, volatile media, and transmission media. Non-volatile media includes, for example, optical or magnetic disks, such as the storage device 1010. Volatile media includes dynamic memory, such as the memory 1006. Transmission media includes coaxial cables, copper wire, and fiber optics, including the wires that comprise the bus 1002.
[0146] Common forms of computer-readable media or computer program products include, for example, a floppy disk, a flexible disk, hard disk, magnetic tape, or any other magnetic medium, a CD-ROM, digital video disc (DVD) , a Blu-ray Disc, any other optical medium, a thumb drive, a memory card, a RAM, PROM, and EPROM, a FLASH-EPROM, any other memory chip or cartridge, or any other tangible medium from which a computer can read.
[0147] Various forms of computer readable media may be involved in carrying one or more sequences of one or more instructions to the processor 1004 for execution. For example, the instructions may initially be carried on the magnetic disk of a remote computer. The remote computer can load the instructions into its dynamic memory and send the instructions over a telephone line using a modem. A modem local to the computer system 1000 can receive the data on the telephone line and use an infra-red transmitter to convert the data to an infra-red signal. An infra-red detector coupled to the bus 1002 can receive the data carried in the infra-red signal and place the data on the bus 1002. The bus 1002 carries the data to the memory 1006, from which the processor 1004 retrieves and executes the instructions. The instructions received by the memory 1006 may optionally be stored on the storage device 1010 either before or after execution by the processor 1004.
[0148] In accordance with various embodiments, instructions configured to be executed by a processor to perform a method are stored on a computer-readable medium. The computer-readable medium can be a device that stores digital information. For example, a computer-readable medium includes a compact disc read-only memory (CD-ROM) as is known in the art for storing software. The computer-readable medium is accessed by a processor suitable for executing instructions configured to be executed.
[0149] The foregoing describes one or more exemplary embodiments of the present disclosure. Other embodiments are within the scope of the attached claims. In some cases, actions or steps recited in the claims can be performed in an order different from that in the embodiments and the desired results can still be achieved. In addition, the processes depicted in the drawings do not necessarily require the shown particular order or consecutive order for achieving the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0150] The systems, apparatuses, modules, or units set forth in the above embodiments may be specifically implemented by a computer chip or entity, or by a product with some function. A typical implementation device is a server system. Certainly, the present disclosure does not exclude that with the development of computer technology in the future, computers that realize the functions of the above-mentioned embodiments may, for example, be personal computers, laptop computers, on-board human-machine interaction devices, cellular phones, camera phones, smart phones, personal digital assistants, media players, navigation devices, e-mail devices, game consoles, tablet computers, wearable devices, or combinations of any of these devices.
[0151] Although one or more embodiments of the present disclosure provide method operating steps as described in the embodiments or flowcharts, they may include more or fewer operating steps based on conventional or non-creative means. The order of steps listed in the embodiments is merely one way among the numerous step performing orders, and does not represent the only performing order. When performed in actual apparatuses or end products, it is possible to perform sequentially or in parallel (for example, in an environment with a parallel processor or multi-threaded processing, and even in a distributed data processing environment) according to the method shown in the embodiments or drawings.
[0152] The terms “comprise” , “include” , or any other variant thereof are intended to encompass non-exclusive inclusion, so that a process, method, product, or device that includes a series of elements not only includes those elements, but also includes other elements that are not expressly listed, or further includes elements inherent to such process, method, product, or device. Without more restrictions, it does not exclude that there are other identical or equivalent elements in the process, method, product, or device that includes the elements. For example, if words such as “first” and “second” are used to represent names, they do not represent any particular order.
[0153] For the convenience of description, the above apparatus, when described, is divided into various modules according to functions that are described separately. Certainly, when implementing one or more embodiments of the present disclosure, the functions of the modules can be implemented in the same one or more software and / or hardware. The modules that achieve the same function can also be implemented by a combination of multiple sub-modules or sub-units, and so on. The apparatus embodiments described above are only schematic. For example, the division of the units is merely a logical functional division. In actual implementation, there may be other division ways. For example, multiple units or components can be combined or integrated into another system, or some features can be omitted or not performed. Additionally, the coupling or direct coupling or communication connection between each other as shown or discussed may be indirect coupling or communication connection through some interfaces, apparatuses or units, and may be in the form of electrical, mechanical, or other types.
[0154] The present disclosure is described with reference to flowcharts and / or block diagrams of the methods, apparatuses (systems) , and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as the combinations of the processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a general-purpose computer, a special-purpose computer, an embedded processor, or a processor of other programmable data processing device to generate a machine, so that the instructions executed by the computer or the processor of the other programmable data processing devices generate an apparatus for implementing the functions specified in one or more processes of the flowcharts and / or one or more blocks of the block diagrams.
[0155] These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing devices to work in a particular manner, so that the instructions stored in the computer-readable memory generate a manufactured product including an instruction apparatus, which implements the functions specified in one or more processes of the flowcharts and / or one or more blocks of the block diagrams. These computer program instructions can also be loaded onto a computer or other programmable data processing devices, so that a series of operating steps are performed on the computer or other programmable devices to generate a computer-implemented processing. Therefore, the instructions executed on the computer or other programmable devices provide steps for implementing the functions specified in one or more processes of the flowcharts and / or one or more blocks of the block diagrams.
[0156] Those skilled in the art should understand that one or more embodiments of the present disclosure may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, one or more embodiments of the present disclosure may take the form of a computer program product implemented on one or more computer-available storage media (including but not limited to disk memory, CD-ROM, optical memory, etc. ) that contain computer-available program codes therein.
[0157] One or more embodiments of the present disclosure may be described in the general context of computer-executable instructions executed by a computer, for example, program modules. Generally, the program modules include routines, programs, objects, assemblies, data structures, etc. that perform particular tasks or implement particular abstract data types. One or more embodiments of the present disclosure can also be practiced in a distributed computing environment where a task is performed by a remote processing device connected through a communication network. In the distributed computing environment, the program modules can reside in local and remote computer storage media including storage devices.
[0158] The same or similar parts among the various embodiments of the present disclosure can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the apparatus embodiments, since they are substantially similar to the method embodiments, their description are relatively simple, and relevant parts can refer to portions of the description of the method embodiments. In the description of the present disclosure, descriptions with reference to the terms “one embodiment” , “some embodiments” , “an example” , “a specific example” , “some examples” , or the like mean that specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present disclosure. In the present disclosure, schematic descriptions of the foregoing terms do not necessarily refer to the same embodiment or example. In addition, the described specific features, structures, materials, or characteristics may be combined in proper manners in any one or more embodiments or examples. In addition, without contradicting each other, those skilled in the art can combine and assemble different embodiments or examples as well as features of different embodiments or examples described in the present disclosure.
[0159] In addition, when used in the present disclosure, the words “herein” , “foregoing” , “following” , “hereinafter” , “hereinabove” and words of similar meanings shall refer to the entirety of the present disclosure but not any particular part of the present disclosure. Moreover, unless otherwise stated expressly or interpreted in other manners in the used context, conditional language for example, “may” , “can” , “for example” , “such as” and the like used herein are usually intended to express that some embodiments include some features, elements, and / or states but other embodiments do not. Therefore, this conditional language is usually not intended to imply that one or more embodiments require the features, elements, and / or states in any manner, or whether include these features, elements, and / or states, or these features, elements, and / or states are performed in any particular embodiment.
[0160] The above descriptions are only embodiments of one or more embodiments of the present disclosure, and are not used to limit one or more embodiments of the present disclosure. For those skilled in the art, one or more embodiments of the present disclosure may have various changes and variations. Any modification, equivalent replacement, improvement, and the like made within the spirit and principle of the present disclosure shall be included within the scope of the claims.
Claims
1.A method for predicting intrinsic parameters of a retention theory equation for a peptide, the retention theory equation defining a retention time as a function of liquid chromatography (LC) parameters and the intrinsic parameters, the method comprising:obtaining an embedding of an amino acid sequence of the peptide;inputting the embedding to a neural network model trained for predicting the intrinsic parameters of the retention theory equation, so as to obtain the intrinsic parameters for the peptide,wherein the neural network model comprises:an encoder configured to encode the embedding to generate an encoded embedding, anda parameter prediction network configured to generate the intrinsic parameters for the peptide based on the encoded embedding.2.The method of claim 1, wherein the parameter prediction network comprises a limit scale module configured to limit an output of the parameter prediction network to working ranges of the intrinsic parameters of the retention theory equation.3.The method of claim 2, wherein the parameter prediction network comprises a plurality of linear activation modules located upstream of the limit scale module.4.The method of claim 1, wherein the retention theory equation comprises a plurality of intrinsic parameters, the parameter prediction network comprises a corresponding plurality of parameter prediction sub-networks arranged in parallel, each of which configured to generate a respective one of the plurality of intrinsic parameters based on the encoded embedding.5.The method of claim 1, wherein the neural network model is trained by a process comprising the following steps:identifying an amino acid sequence and retention time actual values of a sample peptide separated with a plurality of different combinations of LC parameters;obtaining an embedding of the identified amino acid sequence of the sample peptide;inputting the embedding of the identified amino acid sequence of the sample peptide to the neural network model, so as to obtain the intrinsic parameters for the sample peptide;determining retention time predicted values of the sample peptide based on the obtained intrinsic parameters for the sample peptide and the LC parameters for separation of the sample peptide through the retention theory equation; andupdating parameters of the neural network model through a loss function based on the retention time actual values and the retention time predicted values of the sample peptide.6.The method of claim 1, wherein the retention theory equation is derived from a linear solvent strength (LSS) theory equation which is expressed as: wherein k is a retention factor, k0 is the retention factor whenS is a slope constant in the LSS theory equation, is a fraction of a component of a mobile phase in LC.7.The method of claim 6, wherein the intrinsic parameters of the retention theory equation comprise S and coefficients of an expression of k0 or ln k0 with respect to a LC parameter of the retention theory equation.8.The method of claim 6, wherein the LC parameters of the retention theory equation comprise a mobile phase gradient indicating as a function of time t, a column temperature T, and a flow rate F,the intrinsic parameters of the retention theory equation comprise S and coefficients of an expansion of k0 or ln k0 with respect to T.9.A method for predicting a retention time for a peptide, comprising:obtaining intrinsic parameters for the peptide using the method of any of claims 1-8;obtaining liquid chromatography (LC) parameters to be used for separation of the peptide;calculating a retention time for the peptide based on the obtained intrinsic parameters and the obtained LC parameters through the retention theory equation.10.A method for predicting a retention time for a peptide, comprising:obtaining an embedding of an amino acid sequence of the peptide;obtaining liquid chromatography (LC) parameters to be used for separation of the peptide;inputting the embedding and the LC parameters to a neural network model trained for predicting the retention time, so as to obtain the retention time for the peptide,wherein the neural network model comprises:an encoder configured to encode the embedding to generate an encoded embedding, anda prediction head configured to generate the retention time for the peptide based on the encoded embedding and the LC parameters.11.The method of claim 10, wherein the prediction head comprises:a parameter prediction network configured to generate intrinsic parameters of a retention theory equation for the peptide based on the encoded embedding, the retention theory equation defining a retention time as a function of the LC parameters and the intrinsic parameters;a retention time calculation unit configured to calculate a retention time for the peptide based on the predicted intrinsic parameters and the obtained LC parameters through the retention theory equation.12.The method of claim 11, wherein the parameter prediction network comprises a limit scale module configured to limit an output of the parameter prediction network to working ranges of the intrinsic parameters of the retention theory equation.13.The method of claim 12, wherein the parameter prediction network comprises a plurality of linear activation modules located upstream of the limit scale module.14.The method of claim 11, wherein the retention theory equation comprises a plurality of intrinsic parameters, the parameter prediction network comprises a corresponding plurality of parameter prediction sub-networks arranged in parallel, each of which configured to generate a respective one of the plurality of intrinsic parameters based on the encoded embedding.15.The method of claim 11, wherein the retention theory equation is derived from a linear solvent strength (LSS) theory equation which is expressed as: wherein k is a retention factor, k0 is the retention factor whenS is a slope constant in the LSS theory equation, is a fraction of a component of a mobile phase in LC.16.The method of claim 15, wherein the intrinsic parameters of the retention theory equation comprise S and coefficients of an expression of k0 or ln k0 with respect to a LC parameter of the retention theory equation.17.The method of claim 15, wherein the LC parameters of the retention theory equation comprise a mobile phase gradient indicating as a function of time t, a column temperature T, and a flow rate F,the intrinsic parameters of the retention theory equation comprise s and coefficients of an expansion of k0 or ln k0 with respect to T.18.The method of claim 10, wherein the prediction head comprises:convolution layers configured to perform feature extraction on the encoded embedding;a flattened layer located downstream of the convolution layers and configured to convert an output of the convolution layers to a one-dimensional vector; andfully connected layers located downstream of the flattened layer and configured to receive an output of the flattened layer and the LC parameters and output a retention time for the peptide.19.The method of claim 10, wherein the neural network model is trained by a process comprising the following steps:identifying an amino acid sequence and retention time actual values of a sample peptide separated with a plurality of different combinations of LC parameters;obtaining an embedding of the identified amino acid sequence of the sample peptide;inputting the embedding of the identified amino acid sequence of the sample peptide and the LC parameters for separation of the sample peptide to the neural network model, so as to obtain retention time predicted values for the sample peptide; andupdating parameters of the neural network model through a loss function based on the retention time actual values and the retention time predicted values of the sample peptide.20.A method comprising:at a computer system having a display:receiving a first input containing first preliminary ultra-violet (UV) chromatography data and first mass spectrometry (MS) data of a plurality of peptides separated with a first combination of liquid chromatography (LC) parameters, wherein peak areas and widths and amino acid sequences of the plurality of peptides are identified from the first preliminary UV chromatography data and the first MS data;generating a plurality of combinations of LC parameters each having a default mobile phase gradient, other LC parameters than the mobile phase gradient in each of the plurality of combinations of LC parameters corresponding to a respective point in a parameter design space with each dimension corresponding to a respective one of the other LC parameters;for each of the plurality of combinations of LC parameters, predicting retention times for one or more selected peptides of the plurality of peptides using the method of any of claims 9-19, and determining a resolution based on the predicted retention times and the identified peak widths for the one or more selected peptides;selecting one of the plurality of combinations of LC parameters having a highest resolution as a candidate combination of LC parameters;receiving a second input to specify a mobile phase gradient;adjusting the mobile phase gradient in the candidate combination of LC parameters to the specified mobile phase gradient, and reconstructing a UV chromatogram from the first preliminary UV chromatography data based on the adjusted candidate combination of LC parameters; anddisplaying a user interface containing the reconstructed UV chromatogram via the display.21.The method of claim 20, wherein reconstructing a UV chromatogram from the first preliminary UV chromatography data based on the adjusted candidate combination of LC parameters comprises:for the adjusted candidate combination of LC parameters, predicting retention times for the plurality of peptides using the method of any of claims 9-19;generating the reconstructed UV chromatogram based on the identified peak areas and widths and the predicted retention times of the plurality of peptides.22.The method of claim 20, comprising:receiving a third input to confirm the reconstructed UV chromatogram displayed on the user interface,wherein a combination of LC parameters corresponding to the reconstructed UV chromatogram that is confirmed is determined as LC parameters recommended for separation of the plurality of peptides.23.The method of claim 20, wherein determining a resolution for each of the plurality of combinations of LC parameters comprises:for each of the one or more selected peptides, determining a resolution for the peptide based on the predicted retention time and the identified peak width for the peptide;determining a lowest resolution among the resolutions for the one or more selected peptides as the resolution for the combination of LC parameters.24.The method of claim 20, comprising:receiving a fourth input containing second preliminary UV chromatography data of the plurality of peptides separated with a second combination of LC parameters that is different from the first combination of LC parameters,wherein the peak areas and widths of the plurality of peptides are identified from the first and second preliminary UV chromatography data.25.A computer system having a display, comprising:means for performing steps of the method of any of claims 20-24.26.An electronic device, comprising:one or more processors; anda memory coupled to the one or more processors and storing computer-executable instructions which, when executed by the one or more processors, cause the one or more processors to perform the method of any one of claims 1 to 24.27.A non-transitory storage medium storing computer-executable instructions thereon which, when executed by one or more processors, cause the one or more processors to perform the method of any one of claims 1 to 24.28.A computer program product comprising instructions which, when executed by a processor, implement the method of any one of claims 1 to 24.29.A computer program causing a computer to implement the method of any one of claims 1 to 24.