Data Characteristics for this Category
Antibody-Drug Conjugate (ADC) regulatory submissions involve diverse data types and complex structures. Data sources primarily include clinical trial reports (Phase I, II, III), non-clinical study reports (pharmacokinetics, pharmacodynamics, toxicology), manufacturing process validation documents, quality standards and testing methods, and published scientific literature. These data update infrequently, typically submitted incrementally with research progress and regulatory requirements. Document structures for ADC submissions are highly modular, following formats like the Common Technical Document (CTD). This includes Module 1 (administrative and submission information), Module 2 (CTD summaries), Module 3 (quality), Module 4 (non-clinical study reports), and Module 5 (clinical study reports). Specifically, Module 3's manufacturing processes and quality control data, and Module 5's clinical data, often involve complex large molecule characteristic parameters (e.g., antibody glycosylation, drug-to-antibody ratio DAR), small molecule toxin structures, linker stability, and release mechanisms. Fields and units are highly specific. For example, DAR values typically appear as integers or decimals, with units of mol/mol. Antibody concentrations use mg/mL, and toxin loads use µg/mg or µg/mL.
Constraints from these Characteristics on "Citing Sources and Traceability"
The modular structure of ADC submission documents requires citations to precisely point to specific CTD modules, sections, page numbers, or even paragraphs. This ensures traceability accuracy. Infrequent data updates mean cited content remains stable for extended periods. However, updates necessitate strict version control and citation refresh mechanisms. The large volume of complex large molecule and small molecule characteristic parameters requires citations to support precise localization of specific figures, tables, or experimental results. An example is pointing to a specific chromatogram or mass spectrum within a batch analysis report. Furthermore, interdisciplinary data (biology, chemistry, clinical medicine) means the citation system must handle documents in various formats. Examples include PDF clinical reports, Excel analysis data, or Word review documents. Checking the reasonableness of citations requires evaluating whether cited experimental data or arguments support submission conclusions, considering ADC-specific mechanisms (e.g., conjugation stability, toxin release kinetics).
Configuration Guidelines
| Configuration Item | Suggested Value | Rationale for this Value |
|---|---|---|
Chunk size | 500–800 characters | ADC submission document paragraphs are often long, containing detailed descriptions and data. Shorter segments might split semantics, affecting citation completeness. |
Recall count | Top 8–12 entries | Ensures coverage of complex ADC mechanisms and multi-dimensional data, improving recall of relevant information. |
Similarity threshold | 0.75–0.85 | Balances accuracy and recall, preventing irrelevant content from being cited while ensuring critical technical details are not missed. |
Rerank result count | Top 5 entries | Focuses on the most relevant key information, reducing engineer review burden and improving traceability efficiency. |
Citation Metadata Fields | File Path, page number, 图表ID | ADC submission documents have a strong need for precise traceability to specific CTD locations. These fields are fundamental. |
citation Consistency Check | Calibrate by actual measurement | Validates the logical consistency between cited data and submission content, specifically for ADC-specific parameters (e.g., DAR value ranges, stability data). |
Three Common Pitfalls
- Citation responses fail to provide specific filenames or page numbers, preventing quick location of original source material. This occurs when knowledge base construction does not extract and associate sufficiently granular metadata.
- The large language model confuses ADC conjugation efficiency data with pharmacokinetic data from non-ADC drugs when checking citation reasonableness. This happens due to a lack of domain-specific calibration for ADC characteristics in the knowledge base, or insufficient context retrieved to differentiate.
- The system reports
quote type errorwhen citing multiple variables. This occurs when the data type of the cited content does not match the preset variable type, for example, attempting to cite a text paragraph into a numerical variable.
How to Confirm Proper Configuration
- Randomly select 10 completed citation segments from submission documents. Check if the cited file path, page number, and paragraph precisely match the original source material.
- For key ADC technical parameters (e.g., DAR value, linker stability), submit queries and verify that the data cited in the model's response falls within a reasonable range and can be traced back to the corresponding experimental report.
- Simulate queries of varying complexity, including scenarios involving multiple data points and cross-document citations. Observe if the number of returned citation entries aligns with expectations and check for any irrelevant citations.
Values given are common starting points and should be measured against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-21.