ACL Digital Life Sciences Used GenAI to Reshape Scientific Data Summarization

Overview

A leading global pharmaceutical enterprise recognized for advancing healthcare with innovative medicines and cutting-edge therapies sought to streamline the processing and validation of scientific research data, particularly complex drug-disease tables, internally. The objective was to leverage Generative AI (GenAI) to automate data summarization from research documents and generate structured templates, thereby enabling faster and more accurate scientific reviews. ACL Digital Life Sciences stepped in as a strategic technology partner to design and deliver an AI-driven solution tailored to the pharma company’s research validation workflow.

Download Case Study

Challenges

Scientific research documents routinely contain densely formatted tables listing drug names and associated diseases. These tables often vary in structure and complexity, making manual extraction and interpretation time-consuming and prone to errors. The client required a scalable AI solution to automate this process while maintaining scientific rigor and ensuring data traceability. Key pain points included:

Manual Data Extraction

Scientists were spending significant time manually identifying and extracting drug-disease pairs from large volumes of research data

Inconsistency in Formatting

Variability in table formats led to inconsistent data outputs and delayed validation workflows

Lack of Standardized Templates

The absence of a structured format made internal scientific verification cumbersome and inefficient

Solution

ACL Digital Life Sciences engineered a GenAI-powered solution using Azure OpenAI’s GPT-3.5 Turbo model, tailored to accurately extract and summarize table data from scientific literature. Key components of the solution:

AI-Powered Extraction: The GPT-3.5 Turbo model was fine-tuned to identify and extract relevant columns, specifically Drug Names and corresponding Diseases, from research tables.
Custom Data Summarization: The extracted data was automatically formatted into structured, human-readable templates designed for scientific review.
Template Design & Formats: Each template included: Drug Name, Disease, Source Document ID
- Notes (for manual verification)
- Output formats included JSON, Excel, and direct integration with scientific validation tools.
Verification Workflow: Internal scientists reviewed the AI-generated templates, creating a feedback loop to refine model outputs and continuously improve summarization accuracy.