Data Characteristics for this Category
Antibody-Drug Conjugate (ADC) registration documents involve diverse data. Core data sources include Clinical Study Reports (ICH E3 format), Non-clinical Study Reports (GLP/GCP standards), CMC (Chemistry, Manufacturing, and Controls) documentation, and pharmaceutical research data. These documents are typically in PDF, Word, or XML formats, containing extensive structured and unstructured text. For example, clinical study reports detail patient enrollment, dosing regimens, adverse events, and efficacy data, including dose units (mg/kg) and time units (weeks, months). The CMC section covers antibody production processes, conjugation reactions, purification steps, quality control metrics (e.g., purity, aggregate content, drug-to-antibody ratio (DAR)), and stability data. Data update frequency is higher during the development phase and primarily focuses on supplementary documentation during the registration phase.
Constraints on Tool Calling and Plugins from these Characteristics
The complexity and diversity of ADC registration documents impose specific requirements on tool calling and plugins. First, the extensive specialized terminology and abbreviations in the documents demand that plugins possess high-precision text recognition and semantic understanding capabilities. This is especially true when extracting key information from unstructured text, such as identifying specific toxicity types from adverse event descriptions. Second, CMC data includes numerous charts and structured data, requiring tools to parse tabular data and perform unit conversions or consistency checks, for example, verifying the fluctuation range of DAR values across different batches. Third, due to the routine nature of document updates and supplements, plugins need to support version management and incremental updates to ensure the real-time accuracy of knowledge base content. Finally, while not mainstream for registration documents, multi-modal data processing capabilities (e.g., images in some non-clinical studies) can pose challenges for information extraction in certain specific situations.
Configuration Settings
| Configuration Item | Recommended Value | Rationale for this Value |
|---|---|---|
maxContext | 8000 tokens | ADC documents have strong contextual relevance. A sufficiently long context window helps understand the overall safety and efficacy of the drug. |
Chunk size | 500–800 characters | Balances semantic completeness and recall efficiency. This avoids long paragraphs diluting key information and prevents context loss from overly short segments. |
Recall count | Top 10–15 entries | Registration documents often require cross-validation from multiple angles. Increasing the number of recalled items improves relevance coverage. |
Similarity threshold | 0.75–0.85 | Ensures the precision of recalled content, preventing irrelevant or low-relevance text from interfering with judgments on ADC-specific issues. |
Rerank result count | Top 5 entries | After high recall, re-ranking focuses on the most critical pieces of information to improve the quality of the final answer. |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | ADC documents contain many large PDF files. Extending the parsing timeout ensures complete file processing. |
Three Common Pitfalls
- Plugin calls return empty results or only partial information. This occurs when document parsing does not fully extract tabular data or embedded image captions.
- Incorrect identification of adverse event types or confusion of dosage units when processing clinical study reports. This results from the underlying model's insufficient recognition capabilities for specific biomedical terminology and measurement units, or inadequate domain-specific fine-tuning.
- During continuous questioning, the model forgets historical ADC context, leading to incoherent answers. This happens when the session management mechanism does not effectively retain or pass complete historical conversation information.
How to Confirm Proper Configuration
- Select an ADC clinical report containing complex tables and specialized terminology. Upload it and use the plugin to extract key data. Verify the completeness and accuracy of the extracted results.
- For the CMC section of ADCs, test if the plugin can correctly identify and extract key quality attributes such as drug-to-antibody ratio (DAR) values and purity percentages. Validate the reasonableness of the numerical ranges.
- Simulate multi-turn question-and-answer scenarios. Ask questions about ADC safety, efficacy, and manufacturing processes. Observe the model's ability to maintain context across different turns and evaluate the logical consistency of the answers.
The values provided 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.