A multi-source input processing and free energy calculation pipeline method for cyclic peptide molecules

Abstract

The application discloses a multi-source input processing and free energy calculation pipeline method for cyclic peptide molecules. The method isolates input sources according to dataset identification; generates fixed microstate three-dimensional conformations for identification and SMILES in CSV, obtains bond order, closed loop connection and stereochemistry for ligands in PDB or mmCIF, and retains input heavy atom coordinates, and cyclic peptides lacking verified bond order or closed loop connection information do not enter subsequent calculation; performs molecular graph discrimination and filtering in response to cyclic peptide processing instructions; splits the parent ligand set according to integer form net charge after determining the fixed microstate; performs pre-equilibrium, atomic partial charge, ligand network, alchemy transformation, calculation and analysis on each non-single same-charge sub-set. For the determined same-charge ligand pair, the relative hydration free energy separation topology solvent leg and vacuum leg without atomic mapping in the transformation stage can also be established, and the two leg results are paired. The application also discloses corresponding equipment and storage media.

Description

Technical Field

[0001] This invention relates to the fields of computational chemistry, molecular simulation, and computer-aided drug design, and in particular to a computer implementation method, apparatus, and storage medium for converting cyclic peptide molecules in tabular files or protein structure files into free energy calculation tasks and analyzing the calculation results. Background Technology

[0002] Alchemical free energy calculations typically cannot begin directly from the original molecular list. Before calculation, it is necessary to obtain ligand structures with well-defined chemical linkages, usable three-dimensional conformations, determined protonation states, consistent formal and net charges, and compliance with the selected free energy protocol. It is also necessary to generate ligand pairs, computational networks, and specific transformation tasks.

[0003] Existing free energy tools are largely geared towards homologous small molecules. Cyclic peptides combine the characteristics of large molecular weight, numerous rotatable bonds, diverse cyclic forms, and slow conformational exchange. Cyclic peptides can form rings at the head and tail, between the side chain and the main chain, between side chains, or through disulfide bonds. Relying solely on file extensions or molecule names can easily lead to the inclusion of non-cyclic peptides or molecules with incomplete chemical linkages in subsequent calculations.

[0004] Raw data for cyclic peptides may also come from different sources. One type of source is spreadsheet files such as CSV, which typically store ligand identifiers, SMILES, and experimental tags, but lack directly usable 3D coordinates. Another type of source is PDB or mmCIF structure files, which may contain the 3D coordinates of the receptor and binding ligand, but the completeness of ligand bond order, cross-residue cyclic linkages, and stereochemical information varies. In particular, ordinary PDB linker records are often insufficient to reliably represent the complete bond order of cyclic peptides; if the bond order is directly guessed, subsequent protonation, parameterization, and free energy end states may represent incorrect chemical structures.

[0005] Protonation also alters the integer net charge of the molecule. Placing ligands with different net charges within the same relative free energy network creates an alchemical transformation of net charge, increasing the risk of periodic electrostatic finite-size effects and end-state handling. The integer sum of atomic charges is not the same concept as the atomic partial charge used in subsequent force fields; the two need to be handled separately in the pipeline.

[0006] Hybrid relative free energy methods typically require the atomic correspondence between the two ligands to be directly used for topology construction. For some cyclic peptide ligand pairs that are large in size, have different closed-loop relationships, or share a small common core, this atomic correspondence may not be suitable as input for hybrid topology. For already defined ligand pairs, the SepTop method can represent the two ligands separately during the transformation phase without passing the atomic correspondence from the network planning phase to the specific transformation; relative hydration free energy also requires protein-free solvent and vacuum environments.

[0007] Therefore, a complete pipeline for establishing the chemical structure of cyclic peptides is needed. This pipeline is not simply a series of commands, but rather establishes constraint relationships between different stages: the structural origin determines the way chemical information is recovered; the fixed microstates determine the formal net charge; cyclic peptide discrimination determines whether it can enter the set of cyclic peptide parent ligands; the formal net charge determines the subset and the network that can be built; and the explicitly chosen free energy protocol determines the atomic correspondence and how the end-state system is used in specific transformations. Summary of the Invention

[0008] The technical problem to be solved by this invention is: how to convert cyclic peptide inputs from table files or protein structure files with varying degrees of chemical information completeness into a set of executable tasks that are chemically well-defined, type-controlled, have a fixed microstate, consistent formal net charge, and are compatible with the relative free energy protocol, and then perform calculations and result analysis.

[0009] To address the aforementioned technical problems, the present invention provides a pipelined method executed by a computing device, comprising the following steps.

[0010] S101 receives the dataset identifier, cyclic peptide processing instructions, selected free energy protocol, and input source. It constructs a parent ligand set based on the dataset identifier and input source. When constructing each parent ligand set, it only scans the processing directory of that parent ligand set, and does not automatically combine ligands from different experimental or structural sources into the same production dataset.

[0011] S102, obtain and verify the basic chemical structure based on the input type.

[0012] For CSV input, ligand identifiers and SMILES are parsed. For PDB or mmCIF input, receptors and ligands are extracted from the structure file. For cyclic peptide production calculations, bond order and cyclic linkages are obtained from the input file or a verified complete ligand template, and stereochemistry is reconstructed based on the input 3D coordinates. If only PDB linkage records are available and no authoritative bond order source is available, the cyclic peptide is stopped from proceeding to protonation and free energy calculations.

[0013] S103: For each ligand, determine and retain only one fixed protonated microstate. For CSV input, generate a 3D conformation and perform geometry optimization after determining the fixed microstate; for structure file input, retain the input heavy atom coordinates during protonation. Preferably, candidate microstates are generated at pH 7.4, and one microstate is selected according to deterministic rules. This step does not represent constant pH molecular dynamics, nor does it represent a population average of candidate microstates, nor does it force all ligands to become neutral molecules.

[0014] S104, in response to cyclic peptide processing instructions, performs cyclic peptide discrimination and parent ligand set admission control. It identifies amino acid-like backbones, amide bonds, molecular rings involving these structures, free terminals, and disulfide bonds based on the molecular diagram. Only molecules with peptide-like or disulfide bond cyclic closure that meet the cyclic peptide discrimination criteria are included in the cyclic peptide parent ligand set.

[0015] S105, after determining the fixed microstates, sum the atomic charges of each ligand to obtain the net charge in integer form, and then split the parent ligand set into multiple subsets according to this net charge in integer form. The neutral subset can be labeled q0, the subset with positive integer charge can be labeled qpN, and the subset with negative integer charge can be labeled qmN, where N is the absolute value of the charge.

[0016] S106, for each subset of net charges of the same form with at least two ligands, execute the subsequent pipeline, including explicit solvent pre-equilibration, atomic partial charge distribution, ligand network generation, alchemical transformation task generation, calculation, and analysis. Before entering the subsequent pipeline, verify again that the current subset has only one integer form of net charge. Subsets with fewer than two ligands do not generate relative free energy transformation tasks.

[0017] S107 generates either a hybrid topology transformation or a separate topology transformation based on the free energy protocol and ligand pair conditions explicitly selected by the user or upper-level program. The hybrid topology transformation uses the atomic correspondences that meet the conditions for topology construction. For like-charge ligand pairs already determined by the shared ligand network, a separate topology path with relative hydration free energy can be selected; this path does not pass the atomic correspondences used in the network planning phase to the separate topology protocol when creating the specific transformation. Mapping quality evaluation itself does not trigger automatic switching from hybrid to separate topology.

[0018] S108, perform each transformation and analyze the results. For the relative hydration free energy separation topological path, calculate the solvent leg and vacuum leg separately, and pair them according to the same directed ligand pairs and the same repeating numbers. Only when the values ​​of the two legs are finite, the units are consistent, and each leg is unique, is the relative hydration free energy obtained by subtracting the vacuum leg free energy from the solvent leg free energy.

[0019] The present invention also provides an electronic device for performing the above-described methods and a computer-readable storage medium storing a computer program that causes a processor to perform the above-described methods.

[0020] Compared to processing methods that do not distinguish between input sources and the completeness of chemical information, the present invention has at least the following advantages:

[0021] 1. The table input and structure input should each adopt a structure preparation path that is appropriate to their information conditions, so as to avoid treating cyclic peptides with only PDB link records and lacking authoritative bond level as already determined chemical structures.

[0022] 2. Cyclic peptide discrimination is performed before ligands are included in the parent ligand set and free energy tasks are generated, so that molecules that do not meet the cyclic peptide mode conditions are not included in the production dataset of this mode.

[0023] 3. Fixed microstate selection, integer net charge splitting, and atomic partial charge allocation are performed in a clear order to avoid confusing formal net charge with force field partial charge and to avoid generating cross-form net charge edges in the same production network.

[0024] 4. A parent ligand set can be automatically assigned to multiple non-singleton charged subsets to complete pre-balancing, network construction, calculation and analysis respectively, while maintaining the isolation of the results of each subset.

[0025] 5. Provide protein-free relative hydration free energy separation topological realizations for known ligand pairs, so that specific transformations do not depend on the common atomic correspondences required by the mixed topology.

[0026] The above effects are limited to input validity, task formation, and result pairing. This invention does not claim to improve free energy prediction accuracy, shorten convergence time, improve cyclic peptide recognition accuracy, or guarantee that any cyclic peptide ligand pair can complete the calculation. Attached Figure Description

[0027] Figure 1 This is the overall flowchart of the multi-source input and free energy calculation pipeline of the present invention.

[0028] Figure 2 This is a flowchart for recovering chemical information from structural files, selecting immobilized microstates, and determining cyclic peptides and controlling the admission of parent ligand sets.

[0029] Figure 3 A schematic diagram of splitting the parent ligand set into integer net charges and performing subsequent pipelined operations on the child sets.

[0030] Figure 4 This diagram illustrates the relationship between shared ligand networks and two types of transformation paths: hybrid topology and separate topology.

[0031] Figure 5This is a schematic diagram of the solvent leg, vacuum leg, and result pairing of the relative hydration free energy separation topology. Detailed Implementation

[0032] The present invention will now be described in conjunction with the accompanying drawings. The embodiments described are for illustrative purposes only and are not intended to limit the scope of protection to specific software names, directory names, target points, force fields, or parameter values.

[0033] I. Terminology

[0034] The "parent ligand set" is a set of ligands built based on explicit dataset labels and input sources. Different sources can use different dataset labels to build separate parent ligand sets.

[0035] The “subset of net charges of the same form” is a set obtained by dividing the total net charge of all atoms of the ligand into integers after the fixed protonated microstates have been determined. This net charge of integers is different from the continuous partial charge allocated to each atom.

[0036] A "fixed protonated microstate" is a discrete chemical state selected from candidate microstates according to deterministic rules. A fixed microstate only indicates that the state is used in this free energy mission, and does not mean that the only state exists in the real solution.

[0037] "Cyclic peptide discrimination and parent ligand set admission control" (hereinafter referred to as "cyclic peptide mode admission control") refers to the system's judgment based on molecular graph rules when the user explicitly selects the cyclic peptide mode. Only molecules that meet the cyclic peptide discrimination criteria are included in the cyclic peptide parent ligand set, while molecules that do not meet the criteria are prevented from entering the set. The system may also have a small molecule mode, but the admission control is not performed in the small molecule mode, and it is not the main focus of this invention.

[0038] "Atomic correspondence" is an atomic mapping between two ligands used for hybrid topology construction or ligand pair quality evaluation. "Separate topology transformation does not use atomic correspondence" means that the mapping is not used as a transformation input when creating and executing a specific separate topology transformation, but it does not mean that the mapping is not used at all in the ligand network planning stage.

[0039] II. Overall Production Line

[0040] like Figure 1 As shown, the overall pipeline 100 includes input and dataset isolation 110, CSV processing branch 120, structure file processing branch 130, fixed microstate processing 140, cyclic peptide mode admission control 150, parent ligand set construction 160, formal net charge splitting 170, same charge subset execution 180, and result analysis 190.

[0041] CSV processing branch 120 includes parsing ligand identifiers and SMILES, selecting fixed microstates, and generating a 3D conformation. Structure file processing branch 130 includes receptor and ligand extraction, chemical link recovery, stereochemical recovery, and protonation while preserving input coordinates. Neither processing branch is required to be executed entirely in a single run; the system executes the appropriate processing branch based on the actual input and enters the common path at point 160 in the parent ligand set construction.

[0042] Figure 1 Optional endpoint estimation branch 200 is also shown. This processing branch shares some structural preparation steps with the alchemical free energy mainline, but does not perform formal net charge subset network construction and alchemical transformation, and therefore does not belong to the same charge subset assignment path.

[0043] III. CSV Input Implementation Methods

[0044] The CSV file must contain at least a recognizable SMILES column, and may also include ligand IDs and experimental labels. The parser normalizes the ligand IDs to Ligands_ID and the structural representation to SMILES; stable IDs can be generated when the input does not contain IDs.

[0045] For each SMILES, candidate microstates are first generated at a preset pH, and then one microstate is selected according to deterministic rules. Subsequently, one or more 3D conformations are generated using distance geometry methods, optimized with molecular force fields, and can be filtered by relative energy windows. Since the CSV source does not have experimental 3D poses, this processing branch yields newly generated conformations.

[0046] IV. Implementation Methods for PDB / mmCIF Input

[0047] like Figure 2 As shown, the structure file processing branch first distinguishes between receptors and ligands from the protein structure file. For mmCIF files, it reads the atomic pairs and bond orders from the chemical composition bond table to recover the internal covalent bonds of the ligands; then it reads the cross-residue linkages belonging to the ligand in the structural linkage table, including linkages that can form closed rings or disulfide bonds. After recovery, it checks whether all ligand atoms belong to a connected molecule and specifies or reconstructs the stereochemistry based on the input three-dimensional coordinates.

[0048] For regular PDB files, CONECT records can be used as atomic linking clues, but cannot be considered a complete bond-level source on their own. When a cyclic peptide lacks corresponding mmCIF chemical composition information, a chemical composition dictionary record, or a manually verified complete ligand template, the program outputs a blocking status and does not continue protonation production.

[0049] In the fixed microstate processing based on structure origin, a heavy atom mapping is established before and after protonation to verify that the heavy atom elements and connectivity relationships have not been accidentally altered, and the original heavy atom coordinates are copied to the selected microstate. A three-dimensional conformation is only re-embedded when the input itself does not have three-dimensional coordinates. The output can record microstate SMILES, formal net charge, pH, selection method, number of candidates, and flags requiring manual verification.

[0050] V. Implementation Method for Cyclic Peptide Discrimination

[0051] The cyclic peptide discrimination module 150 represents ligands as a molecular diagram composed of atoms and chemical bonds. Within this molecular diagram, the module identifies α-peptide residue anchors, each consisting of a nitrogen atom, an α-carbon atom, and a carbonyl carbon atom bonded sequentially; different residue anchors are distinguished by the carbonyl carbon atom. The module also identifies covalent molecular rings in the molecular diagram and detects amide bonds, free peptide chain ends, and disulfide bonds as rule analysis and diagnostic information.

[0052] In one implementation, a cyclic peptide is defined as one that contains at least three distinct α-peptide residue anchors, and shares at least one common atom with each of the same covalent molecular rings and at least two of the distinct α-peptide residue anchors. The uniformity condition does not limit the type of ring-closing bond; for example, rings formed by head-tail amide bonds, side-chain endamide bonds, disulfide bonds, thioether bonds, or carbon-carbon bonds are all judged according to this condition. The free end of the peptide chain can affect the rule score, but does not negate the cyclic peptide determination when the aforementioned structural conditions are met.

[0053] If a ligand contains fewer than three α-peptide residue anchors, or does not contain a covalent molecular ring involving at least two different α-peptide residue anchors, it is not classified as a cyclic peptide. Molecules containing the corresponding molecular ring and amide bond but not meeting the cyclic peptide criteria can be labeled as cyclic amides; other molecules not meeting the criteria are labeled as non-cyclic peptides. When the molecular record is unreadable, the required bond order is unreliable, or the discrimination process cannot be reliably completed, an unknown discrimination result is generated; in cyclic peptide mode, the molecule is not automatically included in the parent ligand set.

[0054] This discrimination is used for admission control in cyclic peptide mode and is not a statistical classification accuracy claim. Cyclic peptide mode only includes molecules classified as cyclic peptides in the parent ligand set; small molecule mode may include ordinary small molecules in the parent ligand set without performing the aforementioned admission control.

[0055] VI. Formal Net Charge Splitting and Subset Execution

[0056] like Figure 3As shown, each ligand in the parent ligand set 310 already possesses a fixed microstate. The system calculates the sum of the atomic charges of each ligand and forms subsets q0, qpN, and qmN according to their charge values. N can be any positive integer, not limited to 1.

[0057] If a subset has only one ligand, it is recorded that the subset cannot establish a relative free energy edge, and the subsequent production stage is skipped. If the subset has at least two ligands, then the following steps are executed in sequence: explicit solvent pre-equilibrium 331, atomic partial charge distribution 332, ligand network generation 333, transformation task generation 334, free energy calculation 335, and result analysis 336.

[0058] A single parent set run generates a shared run identifier 340. Subsets with the same charge are executed sequentially in a predetermined order, using a complete subset identifier consisting of the parent set identifier and a formal net charge suffix. The alchemical transformation task and free energy results for each subset are stored separately according to the complete subset identifier, the selected free energy protocol, the network planning identifier, and the shared run identifier. If a subset fails, the remaining subsets can be tried, and the failure status is finally summarized; this process does not indicate that the failed subsets produced valid results.

[0059] Explicit solvent preequilibration preferentially begins with the existing three-dimensional conformation, placing the ligand in an explicit aqueous and ionic environment for energy minimization, equilibrium, and molecular dynamics production, with the final frame serving as the subsequent structure. Conformation generation only occurs when the ligand lacks a three-dimensional conformation. Atomic partial charge distribution is performed after formal net charge splitting.

[0060] VII. Ligand Networks and Protocol Selection

[0061] like Figure 4 As shown, the ligand network 400 is generated within the same subset of net charges of the same form. Network generation can employ one or more atom mappers to propose atom correspondences, determine ligand pairs using the selected network planning method, and maintain the edge set obtained from the planning. The system first verifies network connectivity, and then performs hard condition verification on each ligand pair. If the ligand network is not connected, any ligand pair undergoes a change in net charge of the form, or the mapping score is still missing or non-finite after attempts to obtain or calculate it, the generation of subsequent alchemical transformation tasks for that ligand network is stopped; the non-finite value includes non-numerical, positive infinity, and negative infinity. For cases where the finite mapping score is lower than a preset threshold, the proportion of mapped heavy atoms is lower than a preset threshold, or the difference in the number of heavy atoms in the ligand is greater than a preset threshold, these are treated as non-blocking quality diagnostic items; corresponding information is output when diagnostic output is enabled, but the ligand pair is not deleted from the ligand network based on this.

[0062] For the same dataset, planning method, and chemical configuration, hybrid topology path 410 and separate topology path 420 can reuse the same protocol-neutral ligand network, thereby generating transformations of different protocols on the same directed ligand pairs. When generating transformation tasks for specific protocols, the system does not filter out edges of the shared ligand network or rewrite its edge set again due to entering a hybrid or separate topology path. Hybrid topology path 410 passes the required atom correspondences to the hybrid topology protocol; when used for cyclic peptide production calculations, it also verifies the elemental consistency of the mapped atoms and stops generating subsequent alchemical transformation tasks for that ligand network if the elements are inconsistent. Separate topology path 420 obtains ligands A and B from the network edges, but does not pass in the atom correspondences when creating specific transformations, and therefore does not use the elemental consistency of the mapped atoms as an admission condition for separate topology transformations.

[0063] The protocol is explicitly selected by the user or the upper-layer program. This implementation does not require the system to automatically switch to a separate topology after a hybrid topology failure, nor does it imply that ligand pairs can still be automatically determined when no available network mappings are available.

[0064] In a set of optional parameters, the hybrid topology protocol uses 11 λ states; the protein complex leg of the separation topology uses 19 λ states, and the solvent leg uses 27 λ states; both the solvent and vacuum legs of the relative hydration free energy separation topology use this 27-state path. The second and third sampling profiles can use production times of 15 ns and 30 ns per state, respectively, with the third sampling profile only extending the time without changing the λ path. The above values ​​are for illustrative purposes only and do not represent optimal parameters for all cyclic peptides.

[0065] VIII. Topological Implementation Method for Separating Relative Hydration Free Energy

[0066] like Figure 5 As shown, for the identified like-charge ligand pairs A and B in the shared network, solvent leg 430 and vacuum leg 440 are constructed respectively.

[0067] Solvent leg 430 has two end states that contain no protein, and the system includes ligand A, ligand B, and an explicit solvent. Vacuum leg 440 has two end states that contain only ligand A and ligand B, and contain no protein or solvent. The protocol verifies that only one ligand component changes at each end when creating the end states, and verifies that the integer net charges of A and B are the same.

[0068] The vacuum leg employs a protocol unit suitable for aperiodic solvent-free systems, and non-bonded interactions are handled without truncation, eliminating the need for a pressure regulator. The operation and analysis sections can reuse the multi-state sampling and analysis capabilities of the separate topological solvent protocol unit.

[0069] The result pairing module 450 searches for a solvent result and a vacuum result along the same directed edge A→B and the same repeating number. If the result file is missing, unreadable, contains NaN or infinity, has different units for the estimated value and uncertainty, has different units for the two legs, or has a repeating leg or inconsistent units across repeats, the pairing is marked as invalid. Valid pairings are calculated using the following formula for relative hydration free energy:

[0070] Formula 1 ΔΔG_hyd(A→B)=ΔG_solvent(A→B)−ΔG_vacuum(A→B).

[0071] If the uncertainties of the two legs are σ_solvent and σ_vacuum respectively, then the combined uncertainty can be obtained by taking the square root of the sum of squares:

[0072] Formula 2 σ_hyd=(σ_solvent²+σ_vacuum²)^(1 / 2).

[0073] IX. Dataset Isolation Example

[0074] Taking PCSK9 cyclic peptide data as an example, the source of the experimental activity table can be established as the parent set PCSK9_assay_2021, and the sources of the crystal structures from the two periods can be established as the parent sets PCSK9_struct_2020 and PCSK9_struct_2024, respectively. These three parent sets are established by the user-provided dataset labels and input paths. The program does not automatically identify the source based on the year, nor does it automatically concatenate the experimental table ligands and crystal ligands into a single production set.

[0075] Each parent set is independently split according to its formal net charge after fixing the microstates. When a crystal structure is merely a view of the binding structure of existing ligands, it can be used to preserve or verify the ligand orientation without creating a second ligand node. After splitting, only the like-charged subsets with at least two ligands are included in the network and relative free energy calculations.

[0076] 10. Optional Endpoint Estimation Branch

[0077] In one alternative implementation, the system receives the prepared receptor and ligand structures and performs MM / PBSA and MM / GBSA, or only MM / GBSA, depending on the user's selection. The system builds a protein-ligand complex for each ligand, performs explicit solvent molecular dynamics, generates the complex, receptor, and ligand topologies, converts the complex trajectory into a format readable by the endpoint method program, and outputs GB and PB endpoint estimates and statistical fields.

[0078] This endpoint branch does not perform formal net charge subset network construction, nor does it belong to the alchemical relative free energy calculation. The endpoint output is the endpoint estimate without conformational entropy terms and should not be directly referred to as the true absolute binding free energy. Preferably, it is only used for quantitative comparisons of validated complex binding postures; complex postures obtained solely through geometric translation can be labeled as diagnostic estimates.

[0079] XI. Small molecule compatible and computing devices

[0080] When the user selects the small molecule mode, the cyclic peptide mode admission control can be omitted, and the corresponding fixed microstate, formal net charge splitting, pre-equilibrium, network, free energy calculation, and analysis steps can still be used. Therefore, the same software implementation can be compatible with small molecules, but the cyclic peptide structure entry point and the cyclic peptide discrimination and parent ligand set admission control are still the main technical objects of this invention.

[0081] The computing device that performs the above method can be a workstation, server, or computing cluster. The device includes a processor, memory, and computing resources for molecular simulation. When the program stored in memory is executed by the processor, it completes the input parsing, molecular diagram processing, task generation, simulation invocation, and result analysis described above. Each function can be performed by a single program or by multiple programs working collaboratively according to the aforementioned data and constraint relationships.

[0082] Those skilled in the art can replace the specific file parsing library, conformation generator, force field, atom mapper, network planner, molecular dynamics engine, and free energy analyzer without departing from the above-described technical concept. As long as the functional relationships of multi-source input normalization, cyclic peptide discrimination and parent ligand set admission control, isomorphic net charge splitting after microstate fixation, and the task of adapting the subset formation protocol are still maintained, they all fall within the scope of the present invention.

Claims

1. A pipelined method for multi-source input processing and free energy calculation of cyclic peptide molecules, executed by a computing device, characterized in that, include: Receive dataset identifier, cyclic peptide processing instructions, selected free energy protocol, and input source corresponding to the dataset identifier. The input source is a table file containing ligand identifiers and linear string representations of molecular structures, or a protein structure file containing ligands. Construct mutually isolated parent ligand sets according to the dataset identifier and input source, so that each ligand can only be included in the parent ligand set corresponding to its dataset identifier and input source. When the input source is a table file containing ligand identifiers and linear string representations of molecular structures, the ligand identifiers and linear string representations of molecular structures are parsed, a fixed protonated microstate is selected, and a three-dimensional conformation of the ligand is generated. When the input source is a protein structure file, the ligand is extracted from the protein structure file, and the bond order, closed loop connection, and stereochemical information of the ligand are obtained. A fixed protonated microstate is selected, and the input heavy atom coordinates are retained during the protonation process. When the bond order or closed loop connection information used for cyclic peptide production calculations cannot be obtained, the ligand is stopped from proceeding to subsequent steps. In response to the cyclic peptide processing instruction, cyclic peptide discrimination is performed based on the amino acid-like backbone, amide bond, covalent cyclic relationship, peptide chain terminal state, and disulfide bond connecting sulfur-containing residues in the ligand molecular diagram, and only ligands with peptide-like cyclic closure or disulfide bond cyclic closure that meet the cyclic peptide discrimination conditions are included in the parent ligand set. After the fixed protonated microstates are determined, the atomic-form charges of each ligand are summed to obtain the integer-form net charge, and the parent ligand set is split into multiple subsets of net charges of the same form according to the integer-form net charge; For each subset of net charges of the same form with at least two ligands, before proceeding to the next stage, it is verified that the subset contains only one integer form of net charge, and explicit solvent pre-equilibrium, atomic partial charge distribution, ligand network generation, alchemical transformation task generation, alchemical free energy calculation and result analysis are performed in sequence. For the same net charge quantum set with fewer than two ligands, no relative free energy transformation task is generated.

2. The method according to claim 1, characterized in that, The table file is a CSV file; after parsing the CSV file, a ligand record containing at least Ligands_ID and SMILES is formed. Distance geometry embedding is performed on the selected fixed protonated microstates to generate a three-dimensional conformation. Force field optimization is performed on the generated three-dimensional conformation, and the conformation is filtered according to the conformation energy.

3. The method according to claim 1, characterized in that, The protein structure file is an mmCIF file; obtaining the bond order, closed loop linkage, and stereochemistry information includes: recovering the internal bond order of the ligand based on the chemical component bond record, recovering the cross-residue closed loop linkages inside the ligand based on the structural linkage record, verifying that the recovered ligand is a connected molecule, and reconstructing the stereochemistry based on the input three-dimensional coordinates; for cyclic peptides that only have PDB linkage records and do not have a verified complete ligand template or other authoritative bond order source, their entry into the protonation production step is stopped.

4. The method according to any one of claims 1 or 3, characterized in that, The selection of the fixed protonated microstates includes: generating candidate microstates under a preset pH condition, selecting one of the candidate microstates according to deterministic rules, verifying that the heavy atom elements and connection relationships remain consistent before and after selection, and recording the selected microstate, the net charge in integer form, the preset pH, and the number of candidate microstates; wherein, the preset pH is 7.

4.

5. The method according to claim 1, characterized in that, The cyclic peptide discrimination includes: identifying α-peptide residue anchors in the molecular diagram of the ligand, each α-peptide residue anchor consisting of a nitrogen atom, an α-carbon atom, and a carbonyl carbon atom bonded sequentially; determining the covalent molecular ring in the molecular diagram and detecting the free ends of the peptide chain; when the ligand contains at least three distinct α-peptide residue anchors, and the same covalent molecular ring shares at least two distinct α-peptide residue anchors with at least one common atom, the ligand is determined to be a cyclic peptide, and the free ends of the peptide chain do not reject ligands that meet the aforementioned criteria; molecules that do not meet the aforementioned criteria, including cyclic amides and other non-cyclic peptide molecules, are excluded from the set of parent ligands.

6. The method according to claim 1, characterized in that, The subsets of net charges of the same form are labeled as neutral subsets, positive integer charge subsets, or negative integer charge subsets according to their integer net charges; the integer net charges are determined before the atomic partial charge allocation and are different from the atomic partial charges.

7. The method according to any one of claims 1 or 6, characterized in that, A shared run identifier is generated for each execution of the parent ligand set. The subsequent stages are executed sequentially for each ligand subset with at least two identical net charge components according to a predetermined charge order. The free energy results are stored separately according to the parent ligand set, the selected free energy protocol, the shared run identifier, and the identical net charge subset.

8. The method according to claim 1, characterized in that, Generating the ligand network includes: proposing atomic correspondences for ligand pairs using one or more atom mappers within the same homogeneous net charge subset; determining ligand pairs according to a preset network planning method and maintaining the set of ligand pairs determined by the network planning method; verifying the connectivity of the ligand network and performing hard condition verification on the ligand pairs; stopping the generation of subsequent alchemical transformation tasks for the ligand network when the ligand network is not connected, or any ligand pair undergoes a change in net charge, or its mapping score is still missing or non-finite after attempted calculation; for at least one of the following cases where the mapping score is finite but lower than a preset score threshold, the proportion of mapped heavy atoms is lower than a preset proportion threshold, or the difference in the number of heavy atoms in the ligands is greater than a preset difference threshold, the corresponding case is determined as a non-blocking quality diagnostic item, and the corresponding ligand pair is maintained; when the selected free energy protocol used for cyclic peptide production calculations uses the atomic correspondences to generate specific alchemical transformations, further verifying the elemental consistency of the mapped atoms, and stopping the generation of subsequent alchemical transformation tasks for the ligand network when the elements are inconsistent.

9. The method according to claim 1, characterized in that, When the selected free energy protocol is a separation topology protocol with relative hydration free energy, for the ligand pairs with the same net charge that have been determined in the ligand network, a solvent transformation containing two ligands and an explicit solvent but not containing proteins, and a vacuum transformation containing two ligands but not containing proteins and solvents are generated respectively. Furthermore, the atomic correspondence of the ligand pairs is not used as the input of the separation topology transformation when creating the solvent transformation and the vacuum transformation.

10. The method according to claim 9, characterized in that, The separation topology protocol verifies that each end state does not contain protein, each end state contains only one changed ligand component, and the ligand pair does not undergo an integer net charge change; depending on whether the system contains solvent, a solvent-containing protocol unit and a vacuum protocol unit are selected, the vacuum protocol unit adopts an aperiodic vacuum system and uninterrupted electrostatic interactions, and does not have a pressure regulator.

11. The method according to any one of claims 9 or 10, characterized in that, The results analysis is performed by pairing a solvent transformation result with a vacuum transformation result one by one according to the same directed ligand pairs and the same repeating number; only when both results are finite values, have the same units, and do not have missing or repeating legs, the relative hydration free energy is obtained by subtracting the vacuum transformation free energy from the solvent transformation free energy, and the combined uncertainty is calculated based on the uncertainty of the two transformation results.

12. A device for multi-source input processing and free energy calculation of cyclic peptide molecules, comprising a processor and a memory, wherein the memory stores a computer program, characterized in that, When the computer program is executed by the processor, it causes the processor to perform the method according to any one of claims 1 to 11.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 11.

Images