PolyCrit V2: LLM Pipeline for PolyIC Expansion
PolyCrit V2 describes our current work to build an LLM-assisted literature extraction pipeline for polymer interaction chromatography. The purpose is to move beyond PolyCrit's original focus on liquid chromatography at critical conditions and expand toward PolyIC: a database focused on all polymer interaction chromatography conditions reported in the literature, where polymer, solvent, and stationary-phase interactions control chromatographic behavior.
Project goal: extract experimental conditions from polymer chromatography PDFs, standardize the extracted fields, manually validate the results, and use the verified records to expand PolyIC from critical-condition entries to a database focused on all interaction chromatography conditions.
From PolyCrit to PolyIC
Original PolyCrit Focus
PolyCrit began as a curated resource for liquid chromatography at critical conditions, emphasizing experimentally reported critical mobile-phase compositions and related column conditions.
Expanded PolyIC Scope
PolyIC is intended to include broader interaction chromatography conditions, including critical, adsorption, gradient, temperature-dependent, normal-phase, and reverse-phase systems.
LLM-Assisted Curation
The V2 effort focuses on extracting structured records from unstructured PDF text, tables, captions, and experimental sections using a controlled LLM workflow.
Human-Reviewed Records
LLM output is treated as a draft extraction. Records are checked against the source PDF before being accepted into the PolyIC database.
Pipeline Workflow
Collect Source PDFs
Gather polymer interaction chromatography papers and retain citation metadata for traceability.
Parse Document Content
Extract text, tables, figure captions, and page-level context from PDFs with varied formatting quality.
Extract Candidate Conditions
Use LLM prompts to identify chromatography modes, polymers, solvents, columns, temperatures, flow rates, detectors, and reported condition types.
Standardize Fields
Normalize solvent names, composition units, polymer names, stationary phases, column dimensions, and experimental units into consistent database-ready fields.
Validate Against PDFs
Compare each extracted entry to the original paper and record evidence such as page, table, figure, or sentence location.
Import Verified Records
Add validated records to PolyIC so the database can support search, comparison, and future machine-learning applications.
Data Fields Targeted for Extraction
The pipeline is designed to capture the experimental information needed to compare interaction chromatography conditions across polymer systems.
Why This Matters
Recover Hidden Data
Many useful conditions are buried in older PDFs, scanned documents, dense tables, or brief experimental descriptions that are difficult to curate manually at scale.
Improve Consistency
Standardized extraction helps make polymer, solvent, column, and operating-condition fields comparable across papers and laboratories.
Support Data-Driven Design
A broader PolyIC database can support searchable comparisons, condition recommendation, and machine-learning models for polymer chromatography.