So Many Papers, So Little Time

As researchers in biology, we’ve all been there: spending countless hours of sifting through irrelevant papers. The sheer volume of new publications (around 300 new papers on spatial transcriptomics monthly!) across an “endless” number of journals like Nature, Cell, and Science makes finding the right information incredibly challenging. Even existing AI tools often fall short, providing broken links or off-topic results. We’re left feeling flooded with irrelevant information, desperately needing a better paper search tool

BioFinderLM: An AI-Powered Literature Search Tool for Biology Research

BioFinderLM is an innovative AI Biology Research Search Tool designed to cut through the noise and deliver highly relevant scientific papers directly to you. Our project tackles the frustrations of traditional search methods by leveraging the power of large language models (LLMs) and the extensive database of PubMed.

How BioFinderLM Works: A Seamless Search Experience

The process is simple yet powerful:

1. Your Query, Transformed: You provide a natural language query, such as “computational tool for alignment of spatial transcriptomics”.
2. Smart Boolean Generation: BioFinderLM utilizes an LLM (like Gemini or Llama3.2) to translate your query into a precise Boolean search expression. This ensures that the search on PubMed is highly targeted.
3. PubMed Integration: We then use the Entrez Programming Utilities (E-utilities) to search PubMed, which comprises over 38 million citations for biomedical literature.
4. Intelligent Filtering: The real magic happens here. Another LLM analyzes the papers found on PubMed, determining their relevance to your query and assigning a confidence level (High, Medium, or Low). This crucial step filters out low-confidence papers, ensuring you only see what truly matters.
5. Clear and Concise Output: You receive a refined list of relevant papers, complete with titles, abstracts, relevance scores, confidence levels, and a reason for their classification. The results are also saved to a convenient JSON file for your convenience.

Real-World Results: Finding What You Need, Faster

Imagine searching for “computational tool for alignment in spatial transcriptomics.” BioFinderLM can take your query, generate a precise PubMed search, find papers (e.g., 48 papers in one example), and then classify them with an LLM, ultimately presenting you with highly relevant results (e.g., 19 relevant papers with medium to high confidence).

Our examples demonstrate the effectiveness of BioFinderLM across various biological research questions:

• For “computational tool for subcellular localization in spatial transcriptomics,” BioFinderLM found 11 papers in PubMed, with 4 identified as highly relevant.
• When searching for “computational tool for RNA localization from sequence data,” even with a larger initial pool (max limit 100 papers), BioFinderLM pinpointed 2 highly relevant papers.
• A query for “computational tool for cell cell communication in spatial transcriptomics” yielded 13 papers, with 9 categorized as highly relevant.

With BioFinderLM, you’ll spend less time sifting through irrelevant information and more time on groundbreaking research. Say goodbye to the “404 Not Found” despair and embrace a more efficient and effective way to discover the biological insights you need.