Gene-associated Disease Discovery Powered by Large Language Models

TL;DR

The paper introduces an LLM-powered framework to discover diseases associated with specific genes by automating literature search, summarization, and evidence-based disease identification from PubMed. It emphasizes real-time updates beyond static databases, leveraging PubMed queries, LangChain-based aggregation, and in-context learning to rank diseases by gene relevance. Evaluation on 1,025 Alzheimer's-disease–associated genes demonstrates that expanding the literature retrieval scope improves top-k disease ranking, with a case study illustrating practical interpretability through GPT-4-derived evidence. The work highlights potential clinical utility in rapid, literature-grounded risk assessment and points to future multi-modal data integration and refined ranking strategies. $HR = \frac{\text{NumberOfHits@K}}{\text{GT}}$ is used to quantify ranking performance.

Abstract

The intricate relationship between genetic variation and human diseases has been a focal point of medical research, evidenced by the identification of risk genes regarding specific diseases. The advent of advanced genome sequencing techniques has significantly improved the efficiency and cost-effectiveness of detecting these genetic markers, playing a crucial role in disease diagnosis and forming the basis for clinical decision-making and early risk assessment. To overcome the limitations of existing databases that record disease-gene associations from existing literature, which often lack real-time updates, we propose a novel framework employing Large Language Models (LLMs) for the discovery of diseases associated with specific genes. This framework aims to automate the labor-intensive process of sifting through medical literature for evidence linking genetic variations to diseases, thereby enhancing the efficiency of disease identification. Our approach involves using LLMs to conduct literature searches, summarize relevant findings, and pinpoint diseases related to specific genes. This paper details the development and application of our LLM-powered framework, demonstrating its potential in streamlining the complex process of literature retrieval and summarization to identify diseases associated with specific genetic variations.

Figures (6)

• Figure 1: Depiction of disease discovery process in clinical practice. It begins with the patient (A) visiting a clinic and undergoing genetic sequencing (B). The physician (C) then analyzes the sequencing results to pinpoint suspicious genetic variations. Subsequently, the physician searches databases or medical literature (D) for records pertinent to these specific genes (E). Finally, the potential disease related to these genes is identified. Our framework is designed to automate the labor-intensive steps from (D) to (F).
• Figure 2: Framework of proposed method. The framework starts from specific genes suspicious to cause disease of the patient. Then the PubMed API is leveraged to search literatures regarding these genes by criteria such as relevance or time. Top K papers are then selected and queried based on crafted prompts by LLMs (e.g., GPT-4). During this phase, the content of the literature is analyzed by LLMs. Relevant diseases are identified and ranked through the in-context learning capabilities of Large Language Models LLMs. This process is iterated several times, with diseases being re-ranked based on the frequency of their occurrence in the outputs.
• Figure 3: Distribution of last-referenced years of selected genes in the dataset. The horizontal axis represents the most recent year a specific gene is referenced. The vertical axis represents frequencies.
• Figure 4: Prompt to instruct GPT-4 for disease ranking generation.
• Figure 5: Distribution of rank of Alzheimer's Disease in the output