Workflow Orchestration for Antibody-Drug Conjugate (ADC)

Data Characteristics

Antibody-Drug Conjugate (ADC) clinical trial pre-screening relies on diverse data sources. These primarily include public clinical trial registries (e.g., ClinicalTrials.gov), internal databases from pharmaceutical companies, medical literature databases (e.g., PubMed, Scopus), and patent databases. Data update frequencies vary. ClinicalTrials.gov typically updates weekly. Internal databases might update in real-time based on research and development progress. Document structures are predominantly semi-structured and unstructured. Examples include clinical trial protocols, investigator brochures, patient medical records, and laboratory test reports. Key fields include target proteins (e.g., HER2, Trop-2), conjugate drug names, linker types, toxin molecules, indications, patient inclusion/exclusion criteria (e.g., ECOG score, creatinine clearance in mL/min), adverse events, and efficacy endpoints (e.g., ORR percentage, PFS in months). This data is often scattered across various reports and tables, presenting complex formats.

Constraints from Data Characteristics on Workflow Orchestration

The diversity and complexity of ADC clinical trial pre-screening data impose specific requirements on workflow orchestration. First, the prevalence of semi-structured and unstructured data means workflows must integrate advanced text parsing and entity recognition modules. For example, extracting critical inclusion/exclusion criteria from clinical trial protocols requires precise natural language parsing to identify specific values, units, and conditions. Second, varying data update frequencies demand flexible data synchronization mechanisms within the workflow. This ensures pre-screening results are based on the latest information. For instance, the workflow should trigger an update when ClinicalTrials.gov publishes new trial information. Third, ADC-specific fields, such as linker type and toxin molecule, require specialized standardization and encoding during data processing. This enables subsequent matching and screening. Finally, integrating multi-source heterogeneous data requires robust data cleaning and deduplication capabilities within the workflow. This addresses issues like inconsistent field naming, missing data, or format errors, ensuring data quality.

Configuration Guidelines

Configuration Item	Suggested Value	Rationale
`Data Source Sync Cycle`	`Weekly`	Balances the update frequency of public databases like ClinicalTrials.gov with internal data processing load.
`Text Chunking Strategy`	`By chapter title，Minimum 500 characters`	Ensures semantic completeness of documents like clinical protocols, preventing critical information from being split.
`Entity Recognition Model`	`Based On BERT Medical Entity Recognition Model`	Improves the accuracy of recognizing ADC-related proper nouns like targets, indications, and inclusion/exclusion criteria.
`Recall count`	`20–30 entries`	Controls the computational cost of subsequent re-ranking and filtering while maintaining recall rate.
`Similarity threshold`	`0.78`	Clinical trial matching requires high accuracy; this threshold effectively filters irrelevant results.
`Code Execution Environment`	`Node.js 18.x`	Most bioinformatics processing scripts are compatible and perform stably in this environment.

Common Pitfalls

Log output does not appear in the expected location when executing custom code within a workflow. This typically results from sandbox environment restrictions on standard output stream redirection in the code execution module. Consult the platform documentation for specific logging mechanisms.
The AI model selection dropdown list in the workflow is empty, preventing model configuration. This usually indicates that the AI model service is not correctly configured or authorized, or there is an issue with the model interface address, preventing the platform from retrieving available model lists.
API calls to create a workflow return a 404 error. This suggests an incorrect API endpoint path or that the FastGPT service has not started the corresponding API interface. Verify the correct path and port information in the API documentation.

Verification Steps

Submit a simulated query containing ADC targets, indications, and key inclusion/exclusion criteria. Check if the workflow's output includes relevant and accurate clinical trial information. Verify the extracted field values.
Review workflow logs. Confirm that the data synchronization module runs as scheduled and without significant errors or warnings. Verify that the data source update frequency meets expectations.
Insert debug outputs at intermediate workflow nodes. Verify that text parsing and entity recognition modules correctly extract key information like ECOG scores and PFS from clinical protocol documents. Check that their units and values are correct.
Run a test case with a code execution node. Confirm that custom code runs correctly and its outputs or side effects match expectations. Also, check that logs are correctly collected and viewable.

Note: The values provided are common starting points. Measure them against specific samples to determine optimal settings.

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.