Data Characteristics for this Category
Antibody-Drug Conjugate (ADC) product data originates primarily from clinical trial reports, patent literature, scientific papers, drug inserts, and various bioinformatics databases. This data updates frequently, especially during clinical trial phases where data is generated in real-time. Document structures are complex, often containing large volumes of unstructured text, charts, molecular structures, sequence information, pharmacokinetic data, and toxicology reports. Key fields include antibody sequences, linker chemical structures, toxin molecular formulas, drug-to-antibody ratio (DAR), target information, indications, clinical phase, adverse reactions, and production batches. Common units include molar concentration (nM), dosage (mg/kg), half-life (hours), bioavailability (%), and various toxicity indicators (e.g., IC50, LD50).
Constraints Imposed by these Characteristics on "Tool Calling and Plugins"
The high complexity and multimodal nature of ADC product data place specific requirements on tool calling and plugin configurations. First, unstructured text, charts, and molecular structures require specialized parsing tools, such as Optical Character Recognition (OCR) or molecular structure recognition plugins, to extract key information. Second, the frequency of data updates necessitates tools that support real-time or near real-time API calls to retrieve the latest clinical developments or patent information. Processing multimodal data means that within a single workflow, intelligent routing mechanisms are needed to dispatch different data types to appropriate processing models. For example, molecular structures go to cheminformatics tools, and plain text goes to natural language processing models. Furthermore, field and unit standardization is critical. Plugins must recognize and convert units from different data sources to ensure data consistency, preventing erroneous calculations or analyses due to unit mismatches.
Configuration Guidelines
| Configuration Item | Suggested Value | Rationale for this Value |
|---|---|---|
max_tokens | 4000 | Ensures the ability to process ADC product documents containing detailed experimental data and background descriptions, preventing truncation of critical information. |
tool_choice | auto or {"type": "function", "function": {"name": "extract_adc_info"}} | ADC data processing often requires calling specific tools for molecular structure parsing or pharmacokinetic calculations. auto mode intelligently selects tools, or a specific tool can be explicitly named for improved accuracy. |
file_upload_limit_mb | 100 MB | Clinical reports and patent documents often include high-resolution charts and molecular structure images, resulting in generally large file sizes. |
parse_timeout_seconds | 300 seconds | Processing complex PDF documents, especially ADC reports with numerous charts and tables, requires longer parsing times. |
embedding_model | text-embedding-3-large | Captures the complex biological and chemical characteristics of ADC products, improving the precision of similarity matching. |
similarity_threshold | 0.75 | A higher similarity threshold helps recall more precise results, tailored to the specialized terminology and molecular structure descriptions unique to ADC products. |
Three Common Pitfalls
- Tool call returns null or incomplete values: This often occurs when a plugin fails to correctly parse ADC molecular structures or pharmacokinetic charts, leading to failed extraction of key fields.
- Multimodal data processing workflow interruption: This happens when plain text information is sent to a molecular structure parsing tool, or a molecular structure image is sent to a plain text processing model. The cause is a lack of correct input data type recognition and routing in the workflow orchestration.
- API call timeout: This is observed when external systems calling the FastGPT API experience prolonged unresponsiveness or return a 504 error. This typically happens when processing large ADC clinical trial reports or complex molecular structure computations, and the backend processing time exceeds the default timeout settings of the API gateway or proxy.
How to Verify Correct Configuration
- Upload a typical ADC product insert PDF and verify that key information (e.g., DAR value, target protein, toxin type) is accurately extracted and structured.
- Construct a mixed query containing both plain text questions and molecular structure diagrams, then verify the system correctly routes to the corresponding text Q&A model and molecular structure recognition tool.
- Invoke an ADC consultation scenario involving multiple tool calls via the API, record and analyze API response times, and ensure results are returned within the expected timeframe.
The values provided are common starting points and should be measured against specific 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.