Deployment and Upgrade for Lead Optimization Registration

Data Characteristics

Registration documents for the lead optimization phase primarily include pharmacological activity data, preliminary toxicology reports, pharmacokinetic (ADME) data, and crystal form analysis with stability study reports. Data sources often include internal laboratory instruments, collaborative research institutions, or Contract Research Organizations (CROs). Updates are closely tied to experimental progress, typically occurring in batches at key experimental milestones, such as after completing structural optimization or biological activity screening for a compound series. Document structures are often PDF experimental reports, Excel or CSV raw data tables, and Word summary documents. Fields and units are highly specialized. For example, pharmacological activity data includes IC50 (unit nM) and Ki (unit nM). Toxicology reports involve LD50 (unit mg/kg). Pharmacokinetic data includes Tmax (unit h), Cmax (unit ng/mL), and AUC (unit ng·h/mL).

Constraints on Deployment and Upgrade

The specialized nature and data sensitivity of lead optimization documents require on-premise deployment to ensure data security and compliance. Document update frequency is relatively low, but each update involves a large volume of data, including extensive unstructured text (experimental reports) and structured data (tables). This demands significant resources for document parsing and vectorization. Accurate recognition and contextual association of key numerical fields and their units, such as IC50 and LD50, require the RAG system to possess robust entity recognition and numerical extraction capabilities to prevent misinterpretations due to unit confusion. Furthermore, since the documents span multiple scientific disciplines, knowledge base construction needs detailed hierarchical structuring and tag management to support precise cross-disciplinary retrieval. Offline deployment necessitates pre-pulling and version locking of external dependencies to ensure stable system operation and subsequent upgrades without network connectivity.

Configuration Guidelines

Configuration Item	Recommended Value	Rationale
`UPLOAD_FILE_MAX_SIZE`	`500 MB`	Experimental report PDFs are often large; support for a high single-file upload limit is needed.
`PARSE_FILE_TIMEOUT_SECONDS`	`600 seconds`	Large file parsing is time-consuming; this prevents parsing failures due to timeouts.
`Chunk size`	`800–1200 characters`	Balances contextual coherence for long reports with RAG retrieval efficiency.
`Recall count`	`Top 10 entries`	Increases coverage of initial recall to ensure no critical information is missed.
`Similarity threshold`	`Calibrate by actual measurement`	Adjust based on actual corpus and business needs using a test set to balance recall and precision.
`Rerank result count`	`Top 5 entries`	Selects the most relevant snippets for presentation, addressing complex lead optimization queries.

Common Mistakes

Knowledge base query results lack critical numerical values or units are incorrect: This occurs when professional fields like IC50, LD50, and their corresponding units are not correctly identified during document parsing.
System upgrade fails or cannot start in an offline environment: This happens if not all Docker images were pre-pulled during deployment or external dependencies were not packaged locally.
Uploading a large experimental report PDF file causes the system to become unresponsive for an extended period or report an error: The PARSE_FILE_TIMEOUT_SECONDS parameter is set too low, leading to file parsing timeouts.

Verification Steps

Upload a PDF experimental report containing key numerical values and units, such as IC50 and Tmax. Verify that the knowledge base content accurately extracts these fields and units.
Disconnect the server from the external network and attempt to restart the FastGPT service. Confirm that the service starts normally and the knowledge base functions are available.
Upload a PDF file close to the UPLOAD_FILE_MAX_SIZE limit. Observe the file parsing progress and confirm successful import into the knowledge base.

Note: 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.