1. Basic situation of the project

1. Project name: Tumor margin tissue metabolism and protein spectrum detection and data analysis

2. Project number: JF-FW-2026-D015

3. Project location: Jinfeng Laboratory

4. Procurement method: inquiry

2. Procurement content

3. Business requirements

1. Delivery time: Complete all experiments and data analysis work and complete acceptance within 30 days after signing the contract.

2. Supply/service location: Jinfeng Laboratory, No. 313 Jinyue Road, Chongqing High-tech Zone；

3. Payment method: Settlement will be carried out after the completion report is delivered and the acceptance is passed. After formal settlement, the entire payment will be paid in one go.

4. Total project price limit: The total price quoted shall not be higher than the limit price of 199,810 yuan. The quoted price is a comprehensive lump sum price, including all equipment fees, software fees, installation fees, testing fees, travel expenses, freight, material fees, taxes, etc. The purchaser will not pay other fees separately.

4. Project requirements

Party B shall provide complete technical services covering sample standardization pre-processing, metabolomics and proteomics mass spectrometry detection, full-process quality control, data processing, single-omics and multi-omics joint analysis, feature screening, classification model construction, mechanism analysis and delivery of scientific research-level results.

 1. Sample preprocessing and mass spectrometry detection

Establish a standardized sample reception, quality control and pre-processing process for tumor and resection tissue, and complete metabolite extraction and protein extraction, quantification, enzymatic hydrolysis and peptide purification respectively. The 130 samples should be randomly and evenly distributed to experimental and machine batches to ensure traceability of the entire process.

Use high-resolution, high-sensitivity LC-MS/MS platform for metabolomics and proteomics detection. Metabolomics adopts positive/negative ion mode according to research needs ； Proteomics uses DIA, DDA or quantitative techniques with equivalent performance, and the identification of peptides and proteins should in principle control FDR ≤ 1%.

2. Whole process quality control

Establish a quality control system covering sample processing, liquid phase, ion source, MS1/MS2, identification and quantitative results, and set up internal standards, blank and pooled QC. In principle, set up no less than 1 QC sample for every 8-10 research samples.

Complete system quality control of RAW file integrity, TIC/BPC, retention time, mass error, signal intensity, metabolic characteristic peaks, number of peptide/protein identifications, missing values, QC-CV, correlation, PCA, abnormal samples, batch effects and instrument drift. Abnormal data should be traced back to the cause and re-tested or analyzed as necessary.

 3. Data processing and identification

Metabolomics should complete peak extraction, alignment, denoising, missing value assessment, internal standard/QC correction, normalization and batch effect processing, and combine public databases, MS/MS spectral libraries and standards for metabolite annotation and identification classification.

Proteomics should complete spectrum retrieval, peptide and protein identification, quantification, filtering of low-confidence results, normalization and batch calibration, and provide original and standardized quantitative matrices, software versions, database versions and main analysis parameters.

 4. Single-omics statistics and functional analysis

The following statistical and functional analyzes were carried out for the proteome and metabolome respectively:

(1) Take patients as independent statistical units. Tumor tissue and multiple resection margin tissues of the same patient were analyzed using paired statistics or linear mixed effects model, with patient as a random effect. ； An ordered trend analysis was performed based on incisal distance. If continuous distance records are available, restricted cubic splines, generalized additive models, or change point analysis are further used to identify the molecular boundaries that transition from tumor-like to relatively normal-like.

(2) Carry out sample correlation, PCA, hierarchical clustering, abnormal samples and batch effect assessment. The differential analysis simultaneously reports the effect size, 95% confidence interval, original P value and BH-FDR, and performs necessary multi-factor correction based on information such as age, gender, tumor stage, pathological components, tumor purity, collection time, and testing batch.

(3) Proteome differential abundance analysis can use limma-trend/DEqMS, MSstats, proDA or other methods suitable for data structures to incorporate technical variables such as batches into statistical models. Distinguish between random deletions and non-random deletions due to low abundance ； For proteins that are completely undetected in a certain group, they are not directly interpreted as quantitative fold differences through simple downward interpolation, but are reported separately as "change in detection status".

(4) Carry out single-variable and multi-variable mutual verification of the metabolome, and clarify data transformation, normalization and scaling methods. PLS-DA/OPLS-DA is only used for result interpretation after cross-validation and no less than 1000 permutation tests are passed. ； VIP value must be evaluated jointly with FDR, effect size and resampling stability, and is not used as a separate basis for screening.

(5) Metabolic characteristics combine isotopes, adducts, in-source fragments and other ion families before statistical analysis to avoid repeated counting of the same compound. Report MSI or equivalent annotation confidence level for each metabolite ； Level 1-2 metabolites enter the identified metabolite ORA/MSEA, and low-level annotations are reported separately. The pathway analysis background is limited to the actual detection range of this platform, and unidentified peaks can be analyzed by mummichog/PSEA based on the complete feature table.

(6) Proteome analysis of GO, KEGG, Reactome, Hallmark ORA/GSEA, protein complexes and PPI； Metabolome analysis was carried out on chemical categories, KEGG/Reactome/MetaCyc pathways, pathway topology and metabolic modules. Conduct sensitivity analysis on key results with different difference thresholds, scaling methods, missing value strategies and database versions.

5. Proteome-metabolome joint analysis

After strictly checking the correspondence between proteome and metabolome samples of the same patient and the same tissue, intra-omics normalization is completed respectively, and then view-level scale balancing is performed to avoid a certain omics dominating the joint results due to differences in feature number and variance. Specifically carried out:

(1) Focus on protein-metabolite pairings within the same reaction, the same pathway, and network neighborhoods, conduct robust correlation or partial correlation analysis, and correct for testing batches and clinical pathology covariates； Exploratory full correlation networks must be corrected for multiple testing.

(2) Based on the UniProt EC number, Rhea, Reactome, and KEGG reaction relationships, construct an "enzyme-substrate-product" directed network, combine the change directions of proteins and metabolites, reaction measurement relationships, and pathway positions, and prioritize links in the same direction such as "enzyme increase-substrate decrease-product increase".

(3) Use MOFA2 to carry out unsupervised multi-omics integration, distinguish shared factors and omics-specific factors of proteome and metabolome, report the variance explanation rate of each factor for each omics, and evaluate its relationship with resection distance, pathological status, clinical characteristics and technical batches.

(4) Use sPLS or DIABLO to carry out supervised multi-omics feature screening and establish a streamlined cross-omics feature combination. The number of components, the inter-omics design matrix, and the number of retained features must be tuned within the training fold and compared with the best single-omics model and the clinical baseline model.

(5) When the number of effective patients meets the requirements, Similarity Network Fusion, NMF or Consensus Clustering can be used to carry out candidate classification, and the reliability of the classification can be evaluated through resampling consistency, clustering stability, clinicopathological correlation and external data recurrence.

(6) Finally, a multi-level directed mechanism network of "key proteins/enzymes→substrates/products→metabolic pathways→edge gradients and pathological phenotypes" with evidence sources, change directions, correlation coefficients, FDR, catalytic reaction relationships and database versions is formed.

6. Feature screening and classification model construction

Focusing on the identification of tumor tissue, different resection margin tissues and evaluation of molecular transition states, clinical/pathological baseline models, proteome models, metabolome models and multi-omics models were established respectively. Prefer interpretable regularization methods such as LASSO, Elastic Net, sPLS-DA or DIABLO ； Random forest, SVM and XGBoost are used as nonlinear control models.

All normalization, missing value handling, batching, feature filtering, hyperparameter optimization, and classification threshold selection must be done within the training fold of cross-validation. All tissues of the same patient must enter the same data folder, and it is strictly prohibited to randomly split tissue samples to cause leakage of patient information.

The model adopts nested cross-validation, repeated stratified cross-validation or Bootstrap optimism correction of patient groupings, and is subject to independent cohort, time cohort or external public cohort validation when data conditions permit. Report ROC-AUC and 95% confidence interval, PR-AUC and positivity rate baseline, sensitivity, specificity, MCC, confusion matrix, Brier score, calibration curve, calibration slope and decision curve net benefit.

At the same time, the selection frequency, coefficient direction and stability of candidate features in repeated resampling are reported. Multi-omics models must perform gain comparisons with clinical baselines and the best single-omics model ； Results that have not been verified by an independent cohort are expressed as "candidate signatures" and are not expressed as verified biomarkers.

7. In-depth data mining and mechanism analysis

Based on the scientific research questions of the purchaser, the following hierarchical in-depth data mining is carried out:

(1) Analysis of molecular gradients of resection margins and tumor field effects: Conduct pairing trend, linear mixed effect, spline/GAM and necessary change point analysis for different resection margin distances, identify proteins and metabolites that continue to change, early recovery, delayed recovery and non-monotonic changes, and locate candidate molecular boundaries that transition from tumor-like to relatively normal-like.

(2) Cross-omics potential factor and functional module analysis: combine MOFA2 shared/specific factors, WGCNA modules and metabolic modules to identify the core axis that reflects both protein regulation and metabolic reprogramming； WGCNA, NMF and consensus clustering are only reported when the number of valid patients and stability evaluation meet the requirements.

(3) Clinical and pathological correlation analysis: evaluate the relationship between key factors, modules, and mechanism links and pathological status of resection margins, tumor purity, stroma/inflammation/necrosis ratio, tumor stage, and follow-up outcomes.； Mixed-effects models were used when duplicate tissue from the same patient existed.

(4) Candidate marker grading: Based on the differential effects of this cohort, margin gradient, cross-omics direction consistency, reaction topology, resampling stability, public cohort recurrence and clinicopathological correlation, candidate proteins, metabolites and joint signatures are classified into levels A/B/C.

(5) Mechanism link analysis: focusing on energy metabolism, amino acid metabolism, lipid metabolism, redox, hypoxia, ECM remodeling, invasion and migration, and immune microenvironment, carry out direction consistency, alternative databases and sensitivity verification of different analysis parameters to form a mechanism hypothesis that can be further experimentally verified.

8. Public database and external evidence verification

According to the priority of "same cancer type resection margin cohort>same cancer type tumor-adjacent cancer cohort>same tissue source tumor-normal cohort", carry out multi-level external validation of candidate proteins, metabolites, pathways and multi-omics factors:

(1) Cohort-level recurrence: Search GEO, TCGA/GDC and other public transcriptome data of the same cancer type, and establish comparable groups based on tumor, para-tumor/resection margin and clinical pathological information. Each external cohort was analyzed independently to assess effect direction, significance, and pathway consistency. ； Random effects meta-analysis was performed when there were multiple comparable cohorts. Transcript level results only serve as indirect support for protein mechanisms and do not replace direct verification of proteins or metabolites.

(2) Proteome reproducibility: Search proteome and phosphoryme data of the same cancer species or similar tissues in CPTAC/PDC and PRIDE/ProteomeXchange to verify the reproducibility of candidate protein abundance, pathway activity, protein complexes and functional modules.

(3) Metabolome reproduction: Search metabolome data of tissues of the same cancer type in Metabolomics Workbench and MetaboLights. Taking into account differences in platforms, chromatographic conditions, ion modes, and annotation levels, priority is given to evaluating high-confidence metabolites, chemical classes, and directional consistency at the pathway level, and incomparable peak intensity data are not directly merged.

(4) Single cell and spatial source positioning: Use GEO’s public single cell/spatial transcriptome and Human Protein Atlas data to analyze that candidate enzymes and pathways are mainly derived from tumor cells, fibroblasts, myeloid cells, lymphocytes or other cell types； If spatial data for the same cancer type exist, further evaluate the spatial changes of the candidate mechanism at the tumor-resection margin interface.

(5) Functional dependence and intervenability: Use the gene dependence, expression, proteome and metabolome data of DepMap/CCLE to evaluate the dependence of candidate enzymes or proteins in corresponding cancer cell lines, and combine resources such as Open Targets to evaluate targets - disease evidence, druggability, and existing drugs or clinical candidates.

(6) Reaction and network evidence: Use UniProt/Swiss-Prot, Rhea, Reactome, STRING, BioGRID, IntAct, SIGNOR and OmniPath to verify EC numbers, catalytic reactions, complexes, physical interactions and directed regulatory relationships. Only connections with manual curation, experimental support, or cross-database duplication support are defined as high-confidence mechanism edges.

Public database analysis must record the search date, database and version, inclusion and exclusion criteria, sample size, tissue source, detection platform, preprocessing method and ID mapping rules, and classify the conclusion as "direct recurrence, direction support, pathway support or no recurrence". If published data do not contain protein or metabolite measurements, transcript-level evidence must not be expressed as direct verification of the corresponding molecules.

The overall analysis should result in:

Qualified data → Patient-level pairing and margin gradient analysis → Stable single-omics signal → Shared/specific multi-omics factors → Enzyme-substrate-product directed mechanism chain → Patient-level anti-information leakage model → Public cohort and database external evidence → Hierarchical candidate markers and verifiable mechanism hypotheses.

Establish an evidence matrix for important candidate markers, and form an A/B/C level conclusion based on the effect size of this cohort, pairing/gradient results, cross-omics direction consistency, MOFA2/DIABLO stability, response topology, public data recurrence, and clinicopathological correlation. The mechanism description should clearly distinguish between the direct measurement results of this study, database supporting evidence, and inferential evidence, and avoid directly stating correlation as causation.

 9. Results delivery and technical services.

Party B shall deliver in full:

1. Metabolome and proteome raw mass spectrometry RAW data；

2. Metabolic characteristic peaks, metabolite identification and protein/peptide identification and quantitative results；

3. Original and standardized quantitative matrix；

4. Complete QC and abnormal sample processing results；

5. Single-omics, multi-omics joint analysis and classification model results；

6. All statistical result tables and graph source data；

7. PDF/SVG vector images and ≥300 dpi high-definition TIFF/PNG images；

8. Software, database, key parameters and R/Python scripts or equivalent reproducible analysis process；

9. Complete project analysis report and methodology description.

The analysis plan should be designed based on the actual scientific research problems of the purchaser and should not only provide a fixed template report； After the completion of the project, no less than 3 rounds of personalized analysis, result review or chart optimization services should be provided.

5. Supplier qualification requirements

Meet the qualifications of Article 22 of the "Government Procurement Law of the People's Republic of China":

1. Have the ability to independently bear civil liability.

2. Have good business reputation and sound financial accounting system.

3. Have the necessary equipment and professional technical capabilities to perform the contract.

4. Have a good record of paying taxes and social security funds in accordance with the law.

5. In the three years before participating in government procurement activities, there should be no major illegal records in business activities.

6. Other conditions stipulated by laws and administrative regulations.

6. Inquiry schedule

1. Inquiry release time: 16:00 on August 26, 2026

2. Inquiry deadline: 16:00 on August 31, 2026

Suppliers should encrypt and compress the quotation list and related attachments (which need to be stamped with the official seal and scanned in color) into a zip format before the quotation deadline, and only send the encrypted compressed package to the Jinfeng Laboratory procurement email jfsys-caigou@jflab.ac.cn. The decompression password and project quotation amount must not be entered in the body of the email, otherwise it will be treated as an invalid quotation. The password will be notified separately after the expiration date by the purchaser. Those who do not quote before the quotation deadline will be deemed to have given up automatically. The subject of the email must be filled in strictly in accordance with the format: [Full name of the project] + Supplier's complete unit name + Quotation document. If the subject is filled in incorrectly, causing the purchaser to be unable to identify and classify the email, and the risk of the email being misjudged as spam by the system is borne by the supplier.

7. Quotation document requirements

1. The supplier's quotation document should include the following attachments (see Part 2 for details):

(1) Quotation letter

(2) Quotation list

(3) Legal representative qualification certificate (if the bidder is an authorized representative, he must also provide a legal representative authorization letter)

(4) Statement

(5) Letter of commitment

(6) Business license or other organization certification documents

2. The bidder should prepare a freshly stamped quotation document (PDF) and send it to the mailbox. The quotation document must be clearly visible, otherwise the quotation will be deemed invalid.

3. The quotation document should be produced in accordance with the requirements of the "Purchase Announcement". The quotation document must be signed and sealed by the legal representative of the bidder or his authorized representative at the prescribed signature place. The signature, seal and content should be complete. If there are any omissions, you will bear the risk of being deemed an invalid quotation.；

4. The quotation document should be written clearly and neatly. Any interline insertions, alterations, additions or deletions must be signed or stamped with the personal seal of the bidder’s legal representative or his authorized representative. Quotation documents that are illegible, unclearly expressed, or may lead to non-unique understanding will be deemed invalid.

Attachment.docx