A method and system for evaluating the anti-proliferative potency of candidate anti-tumor compounds based on a six-protein combination

Abstract

The application discloses a kind of candidate anti-tumor compound anti-proliferation efficacy evaluation method and system based on six-protein combination, belong to the field of drug screening and pharmacological evaluation.The application selects and obtains the specific marker combination consisting of TBL1X, NAPA, PLEKHA7, ACSL1, ARL8B, CCDC90B, determines the weight of each protein by elastic network regression algorithm, constructs six-protein comprehensive scoring system, realizes compound anti-proliferation efficacy quantitative score, percentile calibration and activity classification by combining reference compound score distribution.The application sets multidimensional data quality control rules, supports mass spectrometry, immunodetection and other multi-platform quantification, and supports the construction of versionable management data processing system.The application realizes the effect equivalent to whole proteome evaluation with a small amount of protein markers, greatly reduces the detection cost, and is suitable for the efficacy quantitative screening and activity comparison of various anti-tumor candidate compounds.

Description

Technical Field

[0001] This invention relates to the field of drug screening and pharmacological evaluation, specifically to a combination of six protein biomarkers consisting of TBL1X, NAPA, PLEKHA7, ACSL1, ARL8B, and CCDC90B, and a method for evaluating the relative antiproliferative efficacy of candidate compounds based on this combination. This invention also relates to a data processing system for detecting the corresponding six proteins. Background Technology

[0002] The in vitro potency of candidate antitumor compounds is typically evaluated using methods such as cell viability, cell counting, nucleic acid synthesis, or colony formation. These methods can reflect changes in cell number or proliferation status under certain treatment conditions, but the information obtained is mainly focused on the endpoint phenotype, making it difficult to simultaneously reflect the molecular response state formed after drug treatment, and also making it difficult to form reusable molecular scores across different compound categories, detection platforms, and experimental batches.

[0003] Whole proteomics can comprehensively characterize cellular changes following drug treatment at the molecular level. However, detecting thousands of proteins in routine compound screening typically requires high sample sizes, and incurs significant testing and data processing costs, hindering the development of standardized targeted detection methods, kits, and automated evaluation systems. Therefore, it is necessary to obtain a small set of biomarkers from whole proteomics that retain key efficacy information, and to develop evaluation methods based on these biomarkers that enable quantification, calibration, and quality control.

[0004] Furthermore, in drug perturbation data, compounds with the same active pharmaceutical ingredient, highly similar structures, and the same primary target often exhibit strong correlations. If only ordinary random partitioning is used for model training and validation, the model's generalization ability for unknown compounds may be overestimated. Therefore, it is necessary to evaluate the applicability of biomarker combinations to untrained compound categories by retaining out active pharmaceutical ingredients, chemical structural clusters, or the entire primary target. Summary of the Invention

[0005] This invention aims to compress the high-dimensional efficacy-related proteomics response induced by candidate compounds into a comparable, calibrable, and deployable numerical score using a small number of detectable proteins, thereby solving the problems of high cost of whole proteomics detection, lack of molecular state information in conventional activity endpoints, insufficient generalization across chemical categories, and difficulty in directly comparing results across platforms.

[0006] To solve the above problems, the present invention adopts the following technical solution: A method for evaluating the antiproliferative potency of candidate antitumor compounds based on a six-protein combination includes the following steps: TBL1X, NAPA, PLEKHA7, ACSL1, ARL8B, and CCDC90B were identified as a six-protein combination, and the protein entities, weight directions, and model versions were defined. The abundance and variation of six proteins were quantified in the candidate antitumor compound treatment group and the matched control group. The abundance variation of each protein was multiplied by its corresponding weight and then summed to obtain the six-protein composite proteomic score (CPS-6).

[0007] The six-protein comprehensive score CPS-6 is compared with the score distribution of the reference compound to obtain the reference percentile. The reference percentile is compared with the preset classification threshold to obtain at least one of the following: relative antiproliferative efficacy, efficacy percentile, activity classification, confidence information, and ranking of candidate compound antiproliferative activity.

[0008] The six proteins TBL1X, NAPA, PLEKHA7, ACSL1, ARL8B, and CCDC90B correspond to UniProt accession numbers O60907, P54920, Q6IQ23, P33121, Q9NVJ2, and Q9GZT6, respectively, or are the same gene product or verified isoforms that can be mapped to the corresponding proteins.

[0009] The candidate antitumor compounds are selected from marketed drugs, natural products, natural product derivatives, structural analogs, pharmaceutically acceptable salts, stereoisomers, prodrugs, and metabolites; the method can also be used to evaluate pharmaceutical compositions containing the above-mentioned candidate antitumor compounds.

[0010] The candidate antitumor compound treatment group and the matched control group used the same solvent conditions, culture time, cell density, sample processing method, detection batch and data processing procedure.

[0011] The abundance change values ​​of the protein were selected from the log2 fold change relative to the matched control group, the relative abundance change value calibrated by the bridged sample, the log2 change value of the ratio of endogenous peptide to isotopic internal standard peptide, the normalized immunoassay signal change value, and the z-score calculated based on the reference cohort.

[0012] The six-protein comprehensive score CPS-6 can be converted into its percentile in the reference compound score distribution; and the six-protein weights, feature centers and scales, reference distribution, classification thresholds, detection platforms, cell types and processing times are saved and called as a unified model version.

[0013] The activity classification includes at least two categories: active, marginal, and inactive. The classification threshold is determined using reference compounds with measured growth inhibition data, through optimization based on the Youden index, preset sensitivity, preset specificity, quantile, or cost sensitivity.

[0014] When all six proteins meet the preset quantitative quality requirements, the treatment group and the control group meet the preset sample quantitative quality requirements, and the detection results of the bridging sample are within the preset allowable range, the official CPS-6 result is output. If any protein fails to obtain a detection result that meets the preset quantitative quality requirements, the abundance change value of that protein is not set to zero, and the corresponding sample is marked as having incomplete coverage of the six proteins.

[0015] Furthermore, the abundance of the six proteins was quantified using TMT isotope labeling, iTRAQ isotope labeling, data-independent acquisition mass spectrometry, label-free quantitative mass spectrometry, parallel reaction monitoring mass spectrometry, selected reaction monitoring mass spectrometry, or a combination thereof.

[0016] The abundance of the six proteins was quantified by immunoblotting, digital protein blotting, enzyme-linked immunosorbent assay (ELISA), protein chip, flow cytometry microsphere detection, aptamer detection, and proximity extension analysis.

[0017] Furthermore, the six-protein comprehensive score CPS-6 is calculated according to the following formula: CPS-6=Σβ i X i Among them, X i β represents the log2 fold change of the i-th protein relative to the control, or the change obtained after quality control standardization. i This represents the protein weight coefficient obtained through model training and corresponding to the model version.

[0018] Furthermore, the protein weighting coefficients are obtained through the following steps: Obtain data on protein abundance changes and corresponding growth inhibition after treatment with the reference compound; Correlation analysis was performed on the protein abundance change data and growth inhibition data after processing and concentration correction to screen candidate proteins; Resilient network regression, bootstrap stability, and random forest variables are used to obtain consensus candidate features from candidate proteins; then, nested cross-validation is used to determine the set of proteins that meet the preset prediction performance requirements within the range of the aforementioned consensus candidate features. The final weighting coefficients of the six proteins were fitted using a discovery cohort.

[0019] Among the weighting coefficients, the weighting coefficients corresponding to TBL1X and PLEKHA7 are negative, while the weighting coefficients corresponding to NAPA, ACSL1, ARL8B and CCDC90B are positive.

[0020] Furthermore, when the model weights for the corresponding platform have not yet been established, the six-protein comprehensive score CPS-6 is replaced by a directionally normalized score, with the formula: S=[ Z(TBL1X) Z(PLEKHA7)+Z(NAPA)+Z(ACSL1)+Z(ARL8B)+Z(CCDC90B)] / 6.

[0021] Furthermore, the method is applied to Jurkat tumor cells, HepG2 tumor cells, primary cells, iPSC-derived cells, organoids, or three-dimensional culture systems, with the treatment time for candidate antitumor compounds being 24 hours.

[0022] A system for evaluating the antiproliferative potency of candidate antitumor compounds based on a six-protein combination includes a data receiving module, a quality control module, a normalization module, a model parameter module, a scoring module, a reference comparison module, and a reporting module.

[0023] Further, the data receiving module is used to receive protein abundance data of TBL1X, NAPA, PLEKHA7, ACSL1, ARL8B, and CCDC90B in the candidate compound treatment group and the matched control group; the quality control module is used to determine whether the protein abundance data meets the preset quality requirements; the normalization module is used to obtain the abundance change values ​​of the six proteins based on the protein abundance data of the candidate compound treatment group and the matched control group; the model parameter module is used to store the weights of the six proteins corresponding to the model version; the scoring module is used to calculate the six-protein comprehensive score CPS-6 based on the abundance change values ​​of the six proteins and the corresponding weight coefficients; the reference comparison module is used to compare the six-protein comprehensive score CPS-6 with the reference compound score distribution or a preset threshold; the reporting module is used to output at least one of the following: relative antiproliferative efficacy, efficacy percentile, activity classification, confidence information, and ranking of candidate compound antiproliferative activity.

[0024] Furthermore, the model parameter module stores at least one of the following: model ID, corresponding model weights, feature center, feature scale, detection platform, cell type, processing time, reference distribution, classification threshold, calibration date, and model hash.

[0025] The generalization ability of the CPS-6 scoring module was evaluated using data from drug active ingredients, structural clusters and major targets, Jurkat proteomics, L1000 transcriptomics and natural products.

[0026] The beneficial effects of this invention are: This invention screens a six-protein biomarker combination consisting of TBL1X, NAPA, PLEKHA7, ACSL1, ARL8B, and CCDC90B from the whole proteome, and establishes CPS-6 based on it. While retaining information related to antiproliferative efficacy, it significantly reduces the number of proteins to be detected, thereby reducing the complexity of detection and data processing.

[0027] CPS-6 showed a significant correlation with measured growth inhibition and was able to distinguish compounds with different anti-proliferative activities. When performing overall retention validation based on active pharmaceutical ingredients, chemical structural clusters, and major targets, the fixed six-protein combination maintained relatively stable evaluation performance, indicating that the combination has a certain degree of generalization ability.

[0028] The six-protein combination exhibited a power-related change trend consistent with the HepG2 discovery cohort in independent Jurkat proteomic data; the analysis results of its corresponding genes in L1000 transcriptomic data further indicate that the power-related features have a certain degree of consistency at the transcriptional level.

[0029] This invention can be used to evaluate the relative antiproliferative efficacy of natural products and structurally similar compounds. Among the 24 natural product data used in this invention, the area under the curve (AUC) of CPS-6 in distinguishing between active and inactive compounds was 0.908, and it can be used to compare the relative efficacy differences between structural analogs.

[0030] This invention establishes a scoring, reference distribution, model version, and quality control method to support the detection of six proteins. This method can be further implemented through targeted mass spectrometry, immunoassay, or other specific protein detection platforms, which is conducive to the formation of standardized detection methods and data processing systems. Attached Figure Description

[0031] Figure 1 This is a schematic diagram illustrating the discovery process of the six-protein combinatorial assembly.

[0032] Figure 2 This is a schematic diagram showing the weight orientation of six proteins and their biological functions.

[0033] Figure 3 This is a schematic diagram of the CPS-6 weighted score calculation and reference percentile calibration process based on the abundance of six proteins.

[0034] Figure 4 The results show the classification performance of antiproliferative efficacy of CPS-6 and whole proteome in the discovery cohort.

[0035] Figure 5 The results validate the generalization performance of the six-protein combination under different grouping and retention conditions.

[0036] Figure 6 This diagram illustrates the cross-cell, cross-dose, and cross-molecular-layer validation of CPS-6.

[0037] Figure 7 Results of CPS-6 application in natural product data.

[0038] Figure 8 This is a schematic diagram of the six-protein quantitative detection process based on targeted mass spectrometry. Detailed Implementation

[0039] Example 1: A method for evaluating the antiproliferative potency of candidate antitumor compounds based on a six-protein combination, comprising the following steps: TBL1X, NAPA, PLEKHA7, ACSL1, ARL8B, and CCDC90B were identified as a six-protein combination, and the protein entities, weight directions, and model versions were defined. The abundance and variation of six proteins were quantified in the candidate antitumor compound treatment group and the matched control group. The abundance variation of each protein was multiplied by its corresponding weight and then summed to obtain the six-protein composite proteomic score (CPS-6).

[0040] The six-protein comprehensive score CPS-6 is compared with the score distribution of the reference compound to obtain the reference percentile. The reference percentile is compared with the preset classification threshold to obtain at least one of the following: relative antiproliferative efficacy, efficacy percentile, activity classification, confidence information, and ranking of candidate compound antiproliferative activity.

[0041] The six proteins TBL1X, NAPA, PLEKHA7, ACSL1, ARL8B, and CCDC90B correspond to UniProt accession numbers O60907, P54920, Q6IQ23, P33121, Q9NVJ2, and Q9GZT6, respectively, or are the same gene product or verified isoforms that can be mapped to the corresponding proteins.

[0042] The candidate antitumor compounds are selected from marketed drugs, natural products, natural product derivatives, structural analogs, pharmaceutically acceptable salts, stereoisomers, prodrugs, and metabolites; the method can also be used to evaluate pharmaceutical compositions containing the above-mentioned candidate antitumor compounds.

[0043] The candidate antitumor compound treatment group and the matched control group used the same solvent conditions, culture time, cell density, sample processing method, detection batch and data processing procedure.

[0044] The abundance change values ​​of the protein were selected from the log2 fold change relative to the matched control group, the relative abundance change value calibrated by the bridged sample, the log2 change value of the ratio of endogenous peptide to isotopic internal standard peptide, the normalized immunoassay signal change value, and the z-score calculated based on the reference cohort.

[0045] The six-protein comprehensive score CPS-6 can be converted into its percentile in the reference compound score distribution; and the six-protein weights, feature centers and scales, reference distribution, classification thresholds, detection platforms, cell types and processing times are saved and called as a unified model version.

[0046] The activity classification includes at least two categories: active, marginal, and inactive. The classification threshold is determined using reference compounds with measured growth inhibition data, through optimization based on the Youden index, preset sensitivity, preset specificity, quantile, or cost sensitivity.

[0047] When all six proteins meet the preset quantitative quality requirements, the treatment group and the control group meet the preset sample quantitative quality requirements, and the detection results of the bridging sample are within the preset allowable range, the official CPS-6 result is output. If any protein fails to obtain a detection result that meets the preset quantitative quality requirements, the abundance change value of that protein is not set to zero, and the corresponding sample is marked as having incomplete coverage of the six proteins.

[0048] The abundance of the six proteins was quantified using TMT isotope labeling, iTRAQ isotope labeling, data-independent acquisition mass spectrometry, label-free quantitative mass spectrometry, parallel reaction monitoring mass spectrometry, selected reaction monitoring mass spectrometry, or a combination thereof.

[0049] The abundance of the six proteins was quantified by immunoblotting, digital immunoblotting, enzyme-linked immunosorbent assay (ELISA), protein microarray, flow cytometry, aptamer detection, and proximity extension analysis.

[0050] The six-protein comprehensive score CPS-6 is calculated according to the following formula: CPS-6=Σβ i X i Among them, X i β represents the log2 fold change of the i-th protein relative to the control, or the change obtained after quality control standardization. i This represents the protein weight coefficient obtained through model training and corresponding to the model version.

[0051] The protein weighting coefficients are obtained through the following steps: Obtain data on protein abundance changes and corresponding growth inhibition after treatment with the reference compound; Correlation analysis was performed on the protein abundance change data and growth inhibition data after processing and concentration correction to screen candidate proteins; Resilient network regression, bootstrap stability, and random forest variables are used to obtain consensus candidate features from candidate proteins; then, nested cross-validation is used to determine the set of proteins that meet the preset prediction performance requirements within the range of the aforementioned consensus candidate features. The final weighting coefficients of the six proteins were fitted using a discovery cohort.

[0052] Among the weighting coefficients, the weighting coefficients corresponding to TBL1X and PLEKHA7 are negative, while the weighting coefficients corresponding to NAPA, ACSL1, ARL8B and CCDC90B are positive.

[0053] Before establishing model weights for the corresponding platform, the six-protein comprehensive score CPS-6 is replaced by a directionally normalized score, with the formula: S=[ Z(TBL1X) Z(PLEKHA7)+Z(NAPA)+Z(ACSL1)+Z(ARL8B)+Z(CCDC90B)] / 6.

[0054] The method is applied to Jurkat tumor cells, HepG2 tumor cells, primary cells, iPSC-derived cells, organoids, or three-dimensional culture systems, with a treatment time of 24 hours for the candidate antitumor compounds.

[0055] A system for evaluating the antiproliferative potency of candidate antitumor compounds based on a six-protein combination includes a data receiving module, a quality control module, a normalization module, a model parameter module, a scoring module, a reference comparison module, and a reporting module.

[0056] The data receiving module is used to receive protein abundance data of TBL1X, NAPA, PLEKHA7, ACSL1, ARL8B, and CCDC90B in the candidate compound treatment group and the matched control group; the quality control module is used to determine whether the protein abundance data meets the preset quality requirements; the normalization module is used to obtain the abundance change values ​​of the six proteins based on the protein abundance data of the candidate compound treatment group and the matched control group; the model parameter module is used to store the weights of the six proteins corresponding to the model version; the scoring module is used to calculate the six-protein comprehensive score CPS-6 based on the abundance change values ​​of the six proteins and the corresponding weight coefficients; the reference comparison module is used to compare the six-protein comprehensive score CPS-6 with the reference compound score distribution or a preset threshold; the reporting module is used to output at least one of the following: relative antiproliferative efficacy, efficacy percentile, activity classification, confidence information, and ranking of candidate compound antiproliferative activity.

[0057] The model parameter module stores at least one of the following: model ID, corresponding model weights, feature center, feature scale, detection platform, cell type, processing time, reference distribution, classification threshold, calibration date, and model hash.

[0058] The generalization ability of the CPS-6 scoring module was evaluated using data from drug active ingredients, structural clusters and major targets, Jurkat proteomics, L1000 transcriptomics and natural products.

[0059] Example 2: Screening of efficacy-related candidate proteins.

[0060] HepG2 cells were treated with 367 antitumor-related compounds for 24 hours to obtain perturbed proteomic data. Quality control of the proteomic data was performed, including limiting the false discovery rate of peptide and protein identification to no more than 1%. Missing protein rates were screened based on effective protein quantification, and the quantitative quality of samples was evaluated by combining the number of quantified proteins, reporter ion intensity, and overall intensity distribution for each sample. Simultaneously, cross-batch quantitative consistency was evaluated using common bridging samples from different TMT-labeled batches, and samples or batches that significantly deviated from the preset quality requirements were reviewed or excluded. The median coefficients of variation for technical and biological replicates were 0.03 and 0.04, respectively.

[0061] After the above quality control, a total of 6682 high-confidence proteins were retained for subsequent analysis. Partial Spearman correlation analysis was performed on the log2 fold change of each protein relative to the matched control and the measured growth inhibition rate, after drug treatment concentration correction, to analyze the association between protein expression changes and cell growth inhibition effects. Using q < 0.05 and |ρ| > 0.30 as screening criteria, a total of 549 candidate proteins significantly associated with growth inhibition were obtained, of which 191 were positively correlated with growth inhibition and 358 were negatively correlated.

[0062] Example 3: Determination of the combination of six protein biomarkers.

[0063] Using growth inhibition rate as the response variable, candidate protein abundance change as the predictor variable, and log10 drug treatment concentration as an unpenalized covariate, a resilient network regression model was employed for feature selection. The stability of feature selection was evaluated using 100 bootstrap resampling iterations, with 80% of the samples randomly selected each time. Simultaneously, random forest regression was used to calculate the variable importance of candidate proteins. Consensus candidate features were obtained by combining the results of resilient network, bootstrap stability, and random forest variable importance analyses.

[0064] Nested five-fold cross-validation was further employed to compare the model performance of different numbers of protein combinations within the consensus candidate features obtained above. Relevance screening and resilient network regression feature selection were performed again within each training fold to avoid test fold information from participating in feature selection. The specific proteins selected from different training folds showed some differences, but the functions involved in the selected proteins mainly focused on processes such as cell cycle, intracellular transport, transcriptional regulation, cytoskeleton, protein homeostasis, and metabolic homeostasis.

[0065] Considering stability, model performance, and protein count, a combination with fewer proteins was selected from those with near-optimal predictive performance. Ultimately, a six-protein biomarker combination consisting of TBL1X, NAPA, PLEKHA7, ACSL1, ARL8B, and CCDC90B was determined. The specific process is as follows: Figure 1 As shown.

[0066] Example 4: Functional information of the six proteins, such as Figure 2 As shown.

[0067] Table 1 Functional information of the six proteins

[0068] Example 5: Calculation of CPS-6.

[0069] A discovery cohort was used to fit the six-protein model, yielding the weights βi for TBL1X, NAPA, PLEKHA7, ACSL1, ARL8B, and CCDC90B. CPS-6 was calculated using the following formula:

[0070] Where Xi represents the input feature of the corresponding protein. If the log2 fold change of the original protein relative to the matched control is used as the input feature during model training, then the test compound will also use the original log2 fold change value. If the log2 fold change value is centered or scale-normalized during model training, then the test sample must be processed using the same centering and scale-normalization methods as during training to obtain the input feature. βi represents the model weight of the corresponding protein. In this embodiment, the weights corresponding to TBL1X and PLEKHA7 are negative, while the weights corresponding to NAPA, ACSL1, ARL8B, and CCDC90B are positive.

[0071] When the input protein abundance variation values ​​are centered or scale-normalized during model building, the same data processing method is used in the actual evaluation process, along with the six model weights corresponding to that processing method. The six protein model weights, feature centers and scales, detection platform, cell type, processing time, reference distribution, and classification threshold are all associated and saved with the corresponding model version to ensure that the test data matches the model parameters.

[0072] In this invention, the detection platform refers to a liquid chromatography-mass spectrometry platform (instrument: Orbitrap Exploris480), through which the abundance and sequence information of proteins are obtained.

[0073] Table 2. Weight values ​​corresponding to the six proteins

[0074] Example 6: Direction-normalized scoring.

[0075] When performing early detection validation, quality control, or when model weights for which corresponding platforms have not yet been established, the abundance variation of the six proteins can be converted into a Z-value relative to a reference sample on the same platform, and the directional normalized score can be calculated according to the following formula:

[0076] The score is calculated based on the correlation between the six proteins and antiproliferative efficacy. The directional normalized score and the CPS-6 score calculated using training weights are different scoring methods, and their reference distributions and classification thresholds are established separately.

[0077] Example 7: Calculation example of directional normalized score.

[0078] The table below provides exemplary inputs for illustrating the calculation steps only and does not correspond to specific drug trial results. It is assumed that the six proteins have been converted to Z-values ​​on the same platform, using directional normalized scoring: Table 3. Six proteins and their corresponding Z-values ​​and directional contributions.

[0079] S=(0.40+0.20+0.35+0.50+0.25+0.10) / 6=0.300 When using the CPS-6 model, the corresponding six protein input values ​​are multiplied by the β value corresponding to the model version, summed, and then the reference percentile is determined based on the score distribution of the corresponding reference compound. The overall flowchart is as follows: Figure 3 As shown.

[0080] Example 8: Antimicrobial susceptibility notes for reference compounds.

[0081] To evaluate the ability of CPS-6 to distinguish compounds with different antiproliferative activities and to establish a CPS-6 reference distribution for interpreting the scores of compounds to be evaluated, this embodiment selects 367 reference compounds. Compounds with available HepG2 drug sensitivity data are annotated, and the classification performance of CPS-6 is evaluated based on these annotations. Simultaneously, a reference distribution is established based on the CPS-6 data of the 367 reference compounds to determine the relative position of the compounds to be evaluated within this distribution. The corresponding CPS-6 predicted category is output according to a pre-set or validated classification threshold (the classification threshold can be determined based on a preset quantile of the reference compound score distribution; alternatively, it can be determined using reference compounds with measured growth inhibition data, optimized based on the Youden index, preset sensitivity, preset specificity, or cost-sensitive optimization).

[0082] Drug response data for HepG2 cells were obtained from publicly available databases such as GDSC1, GDSC2, PRISM, ChEMBL, and PubChem. To improve the consistency of compound matching and drug sensitivity data integration across different databases, the names, synonyms, and other identifying information of reference compounds were standardized before data matching, and drug response records in each database were matched with reference compounds accordingly.

[0083] Quantitative indicators that can be used to characterize drug sensitivity in HepG2 cells were extracted from the above databases, including half-maximal inhibitory concentration (IC50), area under the dose-response curve (AUC), Z-score, log2 fold change in cell viability (LFC), pChEMBL (a standardized drug activity indicator in the ChEMBL database), and growth inhibition rate.

[0084] When multiple available assays for the same compound exist from the same data source, the assay value representing higher activity is retained. Specifically, for IC50, AUC, Z-score, and LFC, the lowest values ​​are retained; for pChEMBL and growth inhibition rate, the highest values ​​are retained. When multiple data sources provide IC50 data for the same compound, the corresponding IC50 values ​​are summed using the geometric mean; when multiple data sources provide AUC data, the corresponding AUC values ​​are summed using the arithmetic mean. The above processing yields quantitative antimicrobial susceptibility information for each reference compound that can be used for antimicrobial susceptibility annotation.

[0085] Based on the standardized and aggregated HepG2 drug susceptibility data, reference compounds with available quantitative drug susceptibility information were annotated and classified into three categories: active, marginal, and inactive.

[0086] When a compound has multiple different types of drug susceptibility indicators, the indicators are prioritized in the following order: IC50, pChEMBL, AUC, LFC, Z-score, and growth inhibition rate. The compound's drug susceptibility annotation category is determined by the highest-priority available indicator. This avoids generating multiple conflicting category labels for the same compound due to different sources or different types of drug susceptibility indicators.

[0087] This embodiment uses the following criteria for classification: compounds with IC50 < 10 μM, pChEMBL ≥ 5.0, AUC < 0.80, LFC < -1.0, Z-score < -1.0, or growth inhibition rate > 50% are labeled as Active. Compounds with 10 μM ≤ IC50 ≤ 30 μM, 4.3 ≤ pChEMBL < 5.0, 0.80 ≤ AUC ≤ 0.95, -1.0 ≤ LFC ≤ -0.5, -1.0 ≤ Z-score ≤ 0, or 20% ≤ growth inhibition rate ≤ 50% are classified as Marginal. Compounds with quantitative drug susceptibility data but that do not meet either the Active or Marginal criteria according to the highest priority indicators are labeled as Inactive. Compounds lacking available quantitative drug susceptibility data and for which a clear drug susceptibility category cannot be obtained according to the above rules are marked as Unknown. Unknown compounds are not included in the classification performance evaluation based on clear drug susceptibility labels, but can be used to establish a CPS-6 reference distribution. According to this rule, among the 367 reference compounds, 238 compounds obtained clear HepG2 drug sensitivity annotations, including 99 active compounds, 40 marginal compounds, and 99 inactive compounds; the remaining compounds without clear drug sensitivity annotations were marked as Unknown.

[0088] It should be noted that the aforementioned Active, Marginal, and Inactive drug susceptibility markers are external reference labels obtained from HepG2 quantitative drug susceptibility data in publicly available databases. They are used to evaluate the relationship between CPS-6 and known antiproliferative activities and are not equivalent to the predicted categories obtained subsequently based on CPS-6 scores or CPS-6 reference percentiles.

[0089] Example 9: Evaluation of the ability of CPS-6 to distinguish between drug susceptibility categories.

[0090] Following the methods described in the preceding embodiments, CPS-6 was obtained for each reference compound. For reference compounds with clearly defined drug sensitivity annotations, the database drug sensitivity annotations were used as external references to evaluate the ability of CPS-6 to distinguish between different drug sensitivity categories. In one embodiment, active and inactive drug sensitivity compounds were selected from compounds with clearly defined drug sensitivity annotations for binary classification analysis, with active drug sensitivity as the positive category and inactive drug sensitivity as the negative category. ROC curves were plotted based on the CPS-6 of each compound, and indices such as AUC, sensitivity, and specificity were calculated to evaluate the distinguishing performance of CPS-6 against active and inactive drug sensitivity compounds. The relationship between CPS-6 and antiproliferative efficacy was evaluated for reference compounds with growth inhibition data or clearly defined drug sensitivity annotations. The results showed that CPS-6 was significantly correlated with the measured growth inhibition rate, with a Spearman correlation coefficient ρ of 0.59 and p of 2.63 × 10⁻⁶. -31 In the binary classification analysis of Active and Inactive classes, the AUC was 0.79, with a 95% confidence interval of 0.72–0.85; at the corresponding classification thresholds, the sensitivity was 0.66 and the specificity was 0.85.

[0091] Under the same reference compounds and classification criteria, CPS-6 was compared with a whole-proteome perturbation index calculated based on 6682 proteins. When used to distinguish between active and inactive compounds, the whole-proteome perturbation index had an AUC of 0.78 and a 95% confidence interval of 0.72–0.85, similar to the classification performance of CPS-6. This result indicates that, under the data conditions of this embodiment, CPS-6, composed of six proteins, can retain key proteome information related to antiproliferative efficacy and achieve similar active and inactive classification performance with significantly fewer detection indicators than the whole-proteome index. Figure 4 As shown.

[0092] Example 10: Establishment of CPS-6 reference distribution.

[0093] In addition to evaluating CPS-6 classification performance using reference compounds with clear drug sensitivity annotations, this embodiment further utilizes 367 reference compounds to establish a CPS-6 reference distribution.

[0094] Specifically, the CPS-6 of 367 reference compounds was calculated, and the resulting CPS-6 were used to construct an empirical reference distribution. The establishment of the reference distribution did not require each reference compound to have a clearly defined database susceptibility label. Therefore, in addition to the 238 compounds with clearly defined susceptibility labels, reference compounds with an susceptibility category of "Unknown" could also be included in the CPS-6 reference distribution.

[0095] For any compound to be evaluated, its CPS-6 is obtained using the same or validated comparable method as that used for reference compounds, and then compared with the reference distribution. The proportion of reference compounds in the reference distribution whose CPS-6 is less than or equal to that of the compound to be evaluated is defined as the CPS-6 reference percentile of that compound. This reference percentile is a relative scoring index based on the CPS-6 reference distribution and is not equivalent to drug sensitivity probability, nor is it equivalent to drug sensitivity annotations obtained from external databases such as GDSC1, GDSC2, PRISM, ChEMBL, or PubChem.

[0096] In one exemplary implementation, compounds to be evaluated can be classified based on CPS-6 reference percentiles: compounds with a CPS-6 reference percentile of at least 75% are classified as CPS-6 Active, compounds with a CPS-6 reference percentile of at least 25% and less than 75% are classified as CPS-6 Marginal, and compounds with a CPS-6 reference percentile of at least 25% are classified as CPS-6 Inactive. The aforementioned 25% and 75% reference percentile boundaries are exemplary classification thresholds, not definitive limits on CPS-6 classification boundaries. Percentile boundaries can be adjusted based on the composition of the reference compounds, model version, target application scenario, expected sensitivity or specificity, and external validation results.

[0097] Example 11: Grouping and retention validation of immobilized hexaproteins.

[0098] To reduce model performance overestimation caused by compounds with similar structures or mechanisms of action being included in both the training and test sets, this embodiment groups compounds according to their active pharmaceutical ingredient, chemical structure cluster, and main target. For chemical structure groups, the Tanimoto similarity between compounds is calculated based on their chemical structure fingerprints, with 0.85 used as the similarity threshold for dividing chemical structure clusters. Each validation iteration uses complete groups as the basis for selection, ensuring that compounds within the same group do not appear simultaneously in both the training and test sets.

[0099] While maintaining the composition of the six proteins TBL1X, NAPA, PLEKHA7, ACSL1, ARL8B, and CCDC90B unchanged, the model coefficients were refitted using only the training group, and the correlation between the model and growth inhibition rate was evaluated in the corresponding retention group. The results showed that the Spearman correlation coefficient ρ was 0.59 when the active drug component was retained as a whole; ρ was 0.60 when the chemical structure cluster was retained as a whole; and ρ was 0.58 when the main target was retained as a whole. Figure 5 As shown.

[0100] The above results indicate that, under the condition of fixed composition of the six proteins and model coefficients being refitted only by training data, the six-protein combination maintained similar efficacy-related performance in the three grouping hold-out validations for active drug components, chemical structures, and major targets, suggesting that the six-protein combination has a certain generalization ability for compound groups that did not participate in model fitting.

[0101] Example 12: Validation of the hexaprotein in Jurkat cells (human acute T-lymphoblastic leukemia cell line).

[0102] The efficacy relevance of the six protein biomarkers was validated using independent Jurkat cell drug treatment proteomic data. This data included 144 compounds treated at concentrations of 1 nM, 10 nM, 100 nM, 1 μM, and 10 μM, with corresponding cell growth inhibition data.

[0103] In this dataset, six proteins—TBL1X, NAPA, PLEKHA7, ACSL1, ARL8B, and CCDC90B—were analyzed individually. The overall trends of TBL1X, NAPA, PLEKHA7, ACSL1, and CCDC90B with increasing compound concentration were largely consistent with the efficacy-related changes of their counterparts in the HepG2 discovery cohort. ARL8B, however, did not exhibit the same overall trend as the HepG2 discovery cohort in this independent Jurkat dataset.

[0104] Keeping the six-protein combination unchanged, a comprehensive score was calculated based on the abundance changes, corresponding directions, and weights of the six proteins (TBL1X, NAPA, PLEKHA7, ACSL1, ARL8B, and CCDC90B). The relationship between this score and the measured growth inhibition rate of Jurkat cells was analyzed. The results showed that the association between the comprehensive score and the growth inhibition rate was weak at 1 nM and 10 nM. The association generally strengthened with increasing treatment concentration, reaching the highest level among the five doses at 10 μM. These results indicate that the expression changes induced by the six proteins are dose-dependent on the anti-proliferative effect of Jurkat cells, and the correlation is more pronounced at higher drug exposure levels.

[0105] Further evaluation using Active and Inactive as binary classification endpoints showed that the AUC of the comprehensive score was 0.82, with a 95% confidence interval of 0.69–0.94, indicating that the six-protein combination has a certain ability to distinguish compounds with different antiproliferative activities in this independent Jurkat dataset. Figure 6 As shown.

[0106] Example 13: Validation of the hexaprotein in L1000 transcriptome data.

[0107] In the LINCS-L1000 (NIH LINCS project high-throughput gene expression dataset) Level 5 data, a total of 88 compounds corresponding to the reference compounds of this invention were matched. The matched transcriptome perturbation data came from various cell types and included multiple perturbation conditions formed by different drug treatment concentrations and treatment times, such as... Figure 6 As shown. Different compounds are used with different cell types and detection conditions, including HepG2, A375, A549, HA1E, HCC515, HT29, MCF7, PC3, and VCAP. Specific cell types, treatment concentrations, and treatment times are based on the corresponding perturbation records matched to each compound in LINCS-L1000.

[0108] Z-scores of the expression of the corresponding genes TBL1X, NAPA, PLEKHA7, ACSL1, ARL8B, and CCDC90B were extracted. To evaluate whether a pre-determined pattern of efficacy of the six biomarkers at the protein level could be observed at the transcriptional level, the orientation and relative weights of each biomarker in the protein-level CPS-6 model were fixed to project the expression changes of the six genes, without refitting the orientations or weights using L1000 data. After weighting and normalization by the sum of the absolute values ​​of the weights, the comprehensive response values ​​of the six genes under each perturbation condition were obtained.

[0109] For multiple sets of gene expression data obtained for the same compound under different cell types, treatment concentrations, and treatment times, the comprehensive response value of six genes was calculated, and the median was taken as the comprehensive response value of the compound at the transcriptional level to reduce the influence of single cell type or single treatment condition on the results. Further correlation analysis was performed between this comprehensive response value and the corresponding measured growth inhibition rate. The results showed that the comprehensive response value of the six genes was positively correlated with the growth inhibition rate, with a Spearman correlation coefficient ρ of 0.38, indicating that under the data conditions of this embodiment, some efficacy-related changes corresponding to the protein level could be observed at the transcriptional level, but the degree of this correlation was limited. Figure 6 As shown.

[0110] It is important to note that, considering the non-one-to-one correspondence between mRNA expression levels and protein abundance, this embodiment follows the direction and relative weights of the protein-level model. The aim is to evaluate the cross-molecular reproducibility of the six biomarker efficacy correlation patterns using pre-determined parameters that are not retrained with L1000 data, rather than to establish a transcriptome version of CPS-6. Therefore, the resulting comprehensive response value of the six genes is only used as an auxiliary validation indicator and is not used as a protein-level CPS-6 indicator, nor is the reference distribution or classification threshold of CPS-6 directly used for efficacy classification.

[0111] Example 14: Application of CPS-6 in natural product data.

[0112] To further evaluate the applicability of CPS-6 to different types of compounds, this embodiment selected 24 structurally diverse natural products for analysis. For natural products lacking publicly available drug susceptibility data but with measured IC50 data from this embodiment, drug susceptibility annotation was performed based on the IC50 obtained after 24 hours of HepG2 cell treatment, following the IC50 classification criteria described in Example 8: IC50 < 10 μM was defined as Active, 10 μM ≤ IC50 ≤ 30 μM as Marginal, and IC50 > 30 μM as Inactive. Of the 24 natural products, 10 were classified as Active, 1 as Marginal, and 13 as Inactive.

[0113] CPS-6 was calculated for each of the aforementioned natural products. In the analysis using Active and Inactive as binary endpoints, the AUC was 0.908, indicating that CPS-6 can effectively distinguish natural products with different antiproliferative activities in the data of this embodiment.

[0114] Eight pairs of structurally similar compounds were further selected for comparison. The results showed that compounds with higher anti-proliferative activity generally had higher CPS-6 levels. Among these, differences in CPS-6 corresponding to differences in anti-proliferative activity were observed among six pairs of structurally similar compounds, such as... Figure 7 As shown in the figure. The above results indicate that CPS-6 can be used not only to evaluate the relative antiproliferative efficacy among different natural products, but also to compare the relative efficacy differences between structural analogs caused by structural changes.

[0115] Example 15: A six-protein quantitative detection scheme based on targeted mass spectrometry (parallel reaction monitoring PRM / multiple reaction monitoring MRM).

[0116] For the six proteins TBL1X, NAPA, PLEKHA7, ACSL1, ARL8B, and CCDC90B, based on the peptide identification results in the discovered proteome, and considering sequence specificity, identification quality, protease cleavage omissions, potential modifications, and chromatographic and mass spectrometric responses, one candidate specific protease digestion peptide was selected for each protein, resulting in a total of six candidate detection peptides. Specific sequences, precursor charges, and retention times are shown in Table 4. In other embodiments, two or more specific peptides may be selected for quantification or auxiliary confirmation of the same target protein.

[0117] HepG2 cells in both the candidate compound treatment group and the matched control group were cultured, treated, and collected under the same conditions. Cells were washed twice with pre-cooled PBS and then lysed in lysis buffer containing a protease inhibitor. In this example, a lysis buffer consisting of 8M urea, 100mM Tris-HCl (pH 8.0), and 1× protease inhibitor was used. Cells were sonicated at 4°C, followed by centrifugation at 14,000×g for 15 min, and the supernatant was collected. Protein concentration was determined using the BCA method. 200 μg of total protein was collected from both the treatment group and the matched control group for subsequent treatment. A bridging sample, obtained by mixing equal volumes of multiple control samples, was used for batch-to-batch quality control.

[0118] Protein samples were added to a final concentration of 10 mM with TCEP and reduced at 55°C for 1 hour. After cooling to room temperature, insulin autoantibody (IAA) was added to a final concentration of 20 mM, and alkylation was performed in the dark for 30 minutes. Urea was diluted to below 1.6 M with 100 mM Tris-HCl (pH 8.0), and sequencing-grade trypsin was added at a protein to trypsin mass ratio of 50:1. Enzymatic digestion was carried out at 37°C for 16 hours. After digestion, the system was adjusted to a pH below 3 using TFA to terminate the reaction. The obtained peptides were desalted using C18 solid-phase extraction, eluted, vacuum-dried, and reconstituted for targeted mass spectrometry detection.

[0119] Stable isotope-labeled internal standard peptides were prepared for each of the six candidate detection peptides. The internal standard peptides can be prepared using peptides containing... 13 C and 15 N-labeled lysine or arginine was used to prepare the working solution, which was then purified and quantified to a concentration of 200 fmol / μL. After desalting the peptides and before mass spectrometry detection, the same volume of internal standard peptide mixture was added to each treatment group and the matched control group. In this example, 10 μL of internal standard peptide mixture was added, and the concentration of the internal standard was adjusted according to preliminary experiments to ensure that the peak area ratio of the endogenous peptide to the corresponding internal standard peptide was preferably within the range of 0.1 to 10.

[0120] PRM was detected using a nano-scale liquid chromatography-mass spectrometry (LC-MS) system. The LC system employed a 75 μm × 5 cm C18 reversed-phase column with a particle size of 2 μm. Mobile phase A was an aqueous solution containing 0.1% formic acid, and mobile phase B was an aqueous solution of 80% acetonitrile containing 0.1% formic acid. The flow rate was 300 nL / min, and the column temperature was 50 °C. Gradient elution conditions were as follows: from 0 min to 50.5 min, mobile phase B linearly increased from 5% to 35%; from 50.5 min to 60 min, mobile phase B linearly increased from 35% to 99%.

[0121] In establishing the PRM method, data-dependent acquisition (DDA) data was first used to determine or assist in confirming the retention time, precursor ion, and product ion information of each target peptide. Subsequently, a PRM-targeted acquisition method was established. Based on the precursor ion mass-to-charge ratio, charge state, and retention time of each candidate peptide, targeted acquisition was performed within the corresponding retention time window. The PRM resolution was set to 17,500, and the automatic gain control (AGC) target value was 2 × 10⁻⁶. 5 The maximum ion implantation time was 100 ms, and fragmentation was performed using a normalized collision energy (NCE) with a center value of 28% and a step size of ±3%. These parameters can be further optimized based on the specific mass spectrometry platform and the target peptide response.

[0122] PRM data were processed using Skyline or instrument-specific software. For each candidate peptide, 3-5 b- or y-type product ions with high response and good specificity were selected for peak area extraction. The peak area ratio of the endogenous peptide to the corresponding stable isotope internal standard peptide was calculated as the relative quantification result. Quality control included a retention time difference of less than 0.2 min between the endogenous peptide and the internal standard peptide, a deviation of less than 30% between the relative abundance of the product ion and the reference spectrum or internal standard peptide, a target peak signal-to-noise ratio greater than 10, a technical repeatability CV of less than 20%, and a batch-to-batch bridging sample peptide abundance CV of less than 25%. Peptides that did not meet the preset quality requirements were marked as quantification failures. When all candidate quantification peptides of a protein failed to meet the quality requirements, the protein was marked as quantification failed, and its abundance change value was not assigned to zero.

[0123] For samples where all six proteins met the quantitative quality requirements, the Log2FC of the candidate compound treatment group relative to the matched control group was further calculated. The Log2FC of the six proteins was converted into model input features using a data preprocessing method consistent with the corresponding model version, and then substituted into the CPS-6 calculation formula described in Example 5 to obtain the CPS-6. This was further compared with the reference compound score distribution of the corresponding model version to obtain the reference percentile, and the corresponding activity classification could be output according to a preset classification threshold. The overall process is as follows: Figure 8 As shown.

[0124] Table 4. Candidate specific peptides, precursor charges, and retention times for the six proteins.

[0125] Example 16: CPS-6 data processing system and model version management.

[0126] The CPS-6 data processing system receives hexagonal protein detection data and corresponding quality control information from the candidate compound treatment group and the matched control group. After quality judgment and standardization, it obtains the abundance change value of hexagonal protein and calls the model parameters corresponding to the detection conditions to calculate CPS-6.

[0127] The system first performs a quality assessment on the input data. When using the targeted mass spectrometry detection method described in Example 15, all six proteins should obtain valid quantitative results, the target peptides should meet the preset requirements for retention time, ion ratio, signal-to-noise ratio, and repeatability, and the bridging samples should be within the allowable deviation range. If any protein fails to obtain a satisfactory quantitative result, its missing value is not set to zero; the system marks the sample as having incomplete coverage of the six proteins and does not output the formal CPS-6 results.

[0128] For data that has passed quality control, the system calculates the log2 fold change value of each protein based on the quantitative values ​​of the treatment group and the matched control group. If the corresponding model directly uses the log2 fold change value as input, it is directly used for scoring; if centering and scale normalization were performed during the model building phase, the transformation is performed according to Xi=(Log2FCi-μi) / σi, where μi and σi are the feature center and feature scale of the i-th protein saved in this model version, respectively. The test samples are processed using the same data processing method as in the model training phase, and μi and σi are not recalculated for individual test samples.

[0129] Each model version can be identified by a model ID. In this embodiment, the information stored associated with the model ID includes the UniProt accession number and feature order of the six proteins, model weights βi, feature centers μi, feature scales σi, detection platform, cell type, processing time, reference compound score distribution, classification threshold, bridging sample reference range, calibration date, software version, and model hash. The system selects and calls the matching model version based on the detection platform, cell type, processing time, and data preprocessing method of the data to be tested, thereby ensuring consistency between input features, model parameters, and reference distribution.

[0130] The following is a simulated operation example. The input and output values ​​in this example are only used to illustrate the operation of the CPS-6 data processing system. Assume that a certain test compound was treated in HepG2 cells for 24 hours, and quantitative data for six proteins were obtained using a detection platform matched to the corresponding model version. After quality assessment, all six proteins met the preset quantitative quality requirements. By comparing the candidate compound treatment group with the matched control group, the Log2FC values ​​for TBL1X, NAPA, PLEKHA7, ACSL1, ARL8B, and CCDC90B were -0.145, 0.098, -0.112, 0.119, 0.201, and 0.158, respectively.

[0131] This simulation example uses a model version that directly uses Log2FC as input features, with the model ID example being "CPS6-HepG2-24H-v1.0", and adopts the six protein weights described in Example 5. Substituting the Log2FC values ​​of the above six proteins into the CPS-6 calculation formula, the CPS-6 value is obtained as 25.29.

[0132] Subsequently, the system compared CPS-6=25.29 with the empirical distribution of CPS-6 for 367 reference compounds associated with the model ID. The reference percentile was calculated based on the proportion of reference compounds in the reference distribution whose CPS-6 was less than or equal to that of the test compound, yielding a reference percentile of 85.6% for this simulated sample. When the corresponding model version pre-defines reference percentiles as follows: ≥75% for CPS-6 Active, >25% and <75% for CPS-6 Marginal, and ≤25% for CPS-6 Inactive, this simulated sample is classified as CPS-6 Active.

[0133] The system ultimately outputs the quality control status of six proteins, along with Log2FC, CPS-6 score of 25.29, reference percentile of 85.6%, CPS-6 Active classification, and the model ID and version information used in this calculation. By saving the model parameters, reference distribution, and model version information corresponding to the scoring results, the system enables traceability of CPS-6 calculation results and consistency management between different testing batches.

[0134] In other embodiments, for the six proteins TBL1X, NAPA, PLEKHA7, ACSL1, ARL8B, and CCDC90B, binding reagents capable of specifically recognizing the corresponding target proteins can also be used for detection, such as antibodies, antibody fragments, aptamers, proximity probes, or other specifically binding molecules. Corresponding detection methods may include Western blotting, digital Western blotting, enzyme-linked immunosorbent assay (ELISA), protein chips, flow cytometry microsphere detection, proximity extension analysis, or aptamer detection.

[0135] When using the above detection methods, appropriate quality control measures can be set according to the specific detection platform, including sample lysis, dilution, total protein quantification, exogenous internal standards, and bridging samples. Bridging samples can be set to negative, intermediate, and positive levels to evaluate the stability of the detection process and the consistency between different batches. For specific binding reagents of the target protein, their specificity, cross-reactivity, linear range, and intra- and inter-batch repeatability can be validated. Specifically, for ARL8B, its cross-reactivity with related proteins such as ARL8A can be evaluated; for TBL1X, its ability to distinguish related homologous proteins or isoforms can be evaluated; for ACSL1, consistent cell lysis and sample processing conditions can be used to reduce the impact of differences in protein extraction efficiency.

[0136] After establishing the corresponding detection platform, the relative abundance changes of six proteins can be calculated based on the detection signals of the candidate compound treatment group and the matched control group. Since the signal scale and dynamic range of different detection platforms may differ, platform conversion or calibration parameters can be established using reference samples or bridging samples. Scoring is then performed after verifying the fit between the six protein detection results and the corresponding CPS-6 model. The weights, normalization parameters, reference distribution, and classification thresholds of the corresponding model can be calibrated according to the detection platform and saved as independent model versions.

[0137] When any target protein fails to meet the preset quantitative quality requirements, its missing detection result is not directly assigned as no change in protein abundance, but the sample is marked as incomplete coverage of six proteins; for cases where no corresponding missing feature processing model has been established, no formal CPS-6 score is output.

Claims

1. A method for evaluating the antiproliferative potency of candidate antitumor compounds based on a six-protein combination, characterized in that, Includes the following steps: TBL1X, NAPA, PLEKHA7, ACSL1, ARL8B, and CCDC90B were identified as a six-protein combination, and the protein entities, weight directions, and model versions were defined. The abundance and variation values ​​of six proteins were quantified in the candidate antitumor compound treatment group and the matched control group. The abundance variation values ​​of each protein were multiplied by their corresponding weights and summed to obtain the six-protein comprehensive score CPS-6. The six-protein comprehensive score CPS-6 was compared with the score distribution of the reference compound to obtain the reference percentile. The reference percentile was compared with the preset classification threshold to obtain at least one of the following: relative antiproliferative efficacy, efficacy percentile, activity classification, confidence information, and ranking of the antiproliferative activity of the candidate compounds. The six proteins TBL1X, NAPA, PLEKHA7, ACSL1, ARL8B and CCDC90B correspond to UniProt accession numbers O60907, P54920, Q6IQ23, P33121, Q9NVJ2 and Q9GZT6 respectively, or are the same gene product or verified isoform that can be mapped to the corresponding protein; The candidate antitumor compounds are selected from marketed drugs, natural products, natural product derivatives, structural analogs, pharmaceutically acceptable salts, stereoisomers, prodrugs, and metabolites; the method can also be used to evaluate pharmaceutical compositions containing the above-mentioned candidate antitumor compounds. The candidate antitumor compound treatment group and the matched control group used the same solvent conditions, culture time, cell density, sample processing method, detection batch and data processing procedure; The protein abundance change values ​​were selected from the log2 fold change relative to the matched control, the relative abundance change value calibrated by the bridged sample, the log2 change value of the ratio of endogenous peptide to isotopic internal standard peptide, the normalized immunoassay signal change value, and the z-score calculated based on the reference cohort. The six-protein comprehensive score CPS-6 can be converted into its percentile in the reference compound score distribution; and the six-protein weights, feature centers and scales, reference distribution, classification thresholds, detection platforms, cell types and processing times are saved and called as a unified model version; The activity classification includes at least two categories: active, boundary, and inactive. The classification threshold is determined by using a reference compound with measured growth inhibition data and optimizing it through Youden index, preset sensitivity, preset specificity, quantile, or cost sensitivity. When all six proteins meet the preset quantitative quality requirements, the treatment group and the control group meet the preset sample quantitative quality requirements, and the detection results of the bridging sample are within the preset allowable range, the official CPS-6 results are output. If any protein fails to obtain a detection result that meets the preset quantitative quality requirements, the abundance change value of that protein will not be set to zero, and the corresponding sample will be marked as having incomplete protein coverage.

2. The method for evaluating the antiproliferative efficacy of candidate antitumor compounds based on a six-protein combination according to claim 1, characterized in that, The abundance of the six proteins was quantified by TMT isotope labeling, iTRAQ isotope labeling, data-independent acquisition mass spectrometry, label-free quantitative mass spectrometry, parallel reaction monitoring mass spectrometry, selected reaction monitoring mass spectrometry, or a combination thereof. The abundance of the six proteins was quantified by immunoblotting, digital protein blotting, enzyme-linked immunosorbent assay (ELISA), protein chip, flow cytometry microsphere detection, aptamer detection, and proximity extension analysis.

3. The method for evaluating the antiproliferative efficacy of candidate antitumor compounds based on a six-protein combination according to claim 1, characterized in that, The six-protein comprehensive score CPS-6 is calculated according to the following formula: CPS-6=Σβ i X i Among them, X i β represents the log2 fold change of the i-th protein relative to the control, or the change obtained after quality control standardization. i This represents the protein weight coefficient obtained through model training and corresponding to the model version.

4. The method for evaluating the antiproliferative efficacy of candidate antitumor compounds based on a six-protein combination according to claim 3, characterized in that, The protein weighting coefficients are obtained through the following steps: Obtain data on protein abundance changes and corresponding growth inhibition after treatment with the reference compound; Correlation analysis was performed on the protein abundance change data and growth inhibition data after processing and concentration correction to screen candidate proteins; Consensus features were obtained from candidate proteins using elastic network regression, bootstrap stability, and random forest variables. Then, nested cross-validation is used to determine the set of proteins that meet the preset prediction performance requirements within the consensus candidate feature range described above; The final weighting coefficients of the six proteins were fitted using a discovery cohort. Among the weighting coefficients, the weighting coefficients corresponding to TBL1X and PLEKHA7 are negative, while the weighting coefficients corresponding to NAPA, ACSL1, ARL8B and CCDC90B are positive.

5. The method for evaluating the antiproliferative efficacy of candidate antitumor compounds based on a six-protein combination according to claim 1, characterized in that, Before establishing model weights for the corresponding platform, the six-protein comprehensive score CPS-6 is replaced by a directionally normalized score, with the formula: S=[ Z(TBL1X) Z(PLEKHA7)+Z(NAPA)+Z(ACSL1)+Z(ARL8B)+Z(CCDC90B)] / 6.

6. The method for evaluating the antiproliferative efficacy of candidate antitumor compounds based on a six-protein combination according to claim 1, characterized in that, The method is applied to human acute T lymphoma cells, human liver tumor cells, primary cells, induced pluripotent stem cell-derived cells, organoids, or three-dimensional culture systems, with a treatment time of 24 hours for candidate antitumor compounds.

7. A system for evaluating the antiproliferative potency of candidate antitumor compounds based on a six-protein combination, characterized in that, It includes a data receiving module, a quality control module, a normalization module, a model parameter module, a scoring module, a reference comparison module, and a reporting module.

8. The antiproliferative efficacy evaluation system for candidate antitumor compounds based on a six-protein combination according to claim 7, characterized in that, The data receiving module is used to receive protein abundance data of TBL1X, NAPA, PLEKHA7, ACSL1, ARL8B, and CCDC90B in the candidate compound treatment group and the matched control group; the quality control module is used to determine whether the protein abundance data meets the preset quality requirements; the normalization module is used to obtain the abundance change values ​​of the six proteins based on the protein abundance data of the candidate compound treatment group and the matched control group; the model parameter module is used to store the weights of the six proteins corresponding to the model version; the scoring module is used to calculate the six-protein comprehensive score CPS-6 based on the abundance change values ​​and the corresponding weight coefficients; the reference comparison module is used to compare the six-protein comprehensive score CPS-6 with the reference compound score distribution or a preset threshold; the reporting module is used to output at least one of the following: relative antiproliferative efficacy, efficacy percentile, activity classification, confidence information, and ranking of candidate compound antiproliferative activity.

9. The antiproliferative efficacy evaluation system for candidate antitumor compounds based on a six-protein combination according to claim 7, characterized in that, The model parameter module stores at least one of the following: model identifier, corresponding model weights, feature center, feature scale, detection platform, cell type, processing time, reference distribution, classification threshold, calibration date, and model hash.

Images