Data Characteristics
Bispecific antibody pharmacovigilance data originates from diverse sources. These include clinical trial reports, real-world evidence (RWE) databases, global adverse drug reaction reporting systems (e.g., WHO VigiBase), and various academic journals and conference abstracts. Data update frequencies vary. Clinical trial data is typically released periodically as research progresses. Real-world data and adverse reaction reports show continuous, dynamic updates.
Document structures often combine structured tables, unstructured text (e.g., medical reports, patient diaries), and semi-structured data (e.g., XML-formatted safety reports). Fields include general patient information, medication history, and adverse event descriptions. They also contain bispecific antibody-specific details like target information, binding modes, and Fc region modification types. Doses are commonly expressed in milligrams (mg) or milligrams per kilogram (mg/kg). Time units include days, weeks, and months. Adverse event severity typically uses CTCAE grading.
Constraints Imposed by Data Characteristics on Tool Calling and Plugins
The diversity of bispecific antibody pharmacovigilance data places specific demands on tool calling and plugins. Structured data requires precise field mapping and data type validation for correct parsing by external tools. Unstructured text needs natural language processing (NLP) capabilities to extract key information, such as detailed adverse event descriptions and timestamps. Semi-structured data requires plugins to handle varying tags and nested structures flexibly.
Due to differing data update frequencies, plugins need to support scheduled or event-driven data synchronization mechanisms for timely information. Bispecific antibody-specific target and molecular structure information requires plugins to call specialized bioinformatics tools for sequence alignment or structural analysis. This aids in determining potential adverse event mechanisms. CTCAE grading for adverse event severity requires plugins to call external medical terminology reference tables or ontology services for standardized encoding.
Configuration Guidelines
| Configuration Item | Recommended Value | Rationale (The values given here are common starting points and should be measured against your own samples.) |
|---|---|---|
maxContext | 3000 Tokens | Covers the detailed description of typical adverse event reports while balancing model processing efficiency. |
parseTimeout | 120 seconds | Handles complex unstructured text processing, such as NLP parsing of lengthy medical reports. |
Recall count (Recall Count) | Top 10 entries (Top 10) | Ensures enough relevant adverse event cases or drug information are retrieved, improving recall rate. |
similarityThreshold | 0.75 | Balances recall and precision, filtering bispecific antibody data highly relevant to the query. |
Rerank result count (Reranked Return Count) | Top 3 entries (Top 3) | Focuses on the most relevant results, reducing model processing burden and improving response speed. |
externalAPIConcurrency | 5 | Accommodates the concurrent processing capabilities of different external bioinformatics tools or database interfaces. |
Common Pitfalls
- Symptom: FastGPT returns "Model stream response is empty, please check if model stream output is normal" after calling an external OpenAPI. Reason: The external API's data format does not match FastGPT's expectations, or it returned a non-standard HTTP status code, preventing FastGPT from parsing it correctly.
- Symptom: Discrepancies arise because the model configured for the application differs from the model configured for knowledge base parameter optimization. Reason: Different models have varying text understanding and generation capabilities. Inconsistent configurations can lead to suboptimal knowledge extraction or answer generation.
- Symptom: Timeout errors or empty results frequently occur when calling bioinformatics tool plugins. Reason: Bispecific antibody-related sequence alignment or structural analysis often involves extensive computations. External tools may exceed FastGPT's configured timeout limits, or input parameters may not meet tool requirements.
Verification Steps
- Simulate typical pharmacovigilance queries. Check if FastGPT's results include correctly extracted key information from structured, unstructured, and semi-structured data. Compare these results against the original data.
- Use FastGPT's log function to track detailed records of external tool calls and plugin executions. Verify that API request parameters, response status codes, and returned content meet expectations and that no abnormal errors exist.
- For queries specific to bispecific antibody targets and molecular structures, verify that plugins successfully called bioinformatics tools and produced correct analysis results (e.g., sequence alignment scores, domain predictions).
- Regularly review the accuracy and timeliness of adverse event report extraction. Adjust parameters like
similarityThresholdandRecall count(Recall Count) by comparing them to human evaluation results until they reach an acceptable business threshold.
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.