Translate 
Abstract
Large language models (LLMs) have been shown to have significant potential in few-shot learning across various fields, even with minimal training data. However, their ability to generalize to unseen tasks in more complex fields, such as biology and medicine has yet to be fully evaluated. LLMs can offer a promising alternative approach for biological inference, particularly in cases where structured data and sample size are limited, by extracting prior knowledge from text corpora. Here we report our proposed few-shot learning approach, which uses LLMs to predict the synergy of drug pairs in rare tissues that lack structured data and features. Our experiments, which involved seven rare tissues from different cancer types, demonstrate that the LLM-based prediction model achieves significant accuracy with very few or zero samples. Our proposed model, the CancerGPT (with ~ 124M parameters), is comparable to the larger fine-tuned GPT-3 model (with ~ 175B parameters). Our research contributes to tackling drug pair synergy prediction in rare tissues with limited data, and also advancing the use of LLMs for biological and medical inference tasks.
Highlights
Large language models (LLMs) have been shown to have significant potential in few-shot learning across various fields, even with minimal training data
Results overview We developed CancerGPT, a few-shot drug pair synergy prediction model for rare tissues
We evaluated the accuracy of our proposed CancerGPT model, other LLM-based models (GPT-2, GPT-3, SciFive20), and general data-driven prediction models (XGBoost, Collaborative Filtering, TabTransformer) (Methods, Fig. 3, 4, Supplementary Figs. 1, and 2, Supplementary Table 2-5) by the area under the precision-recall curve (AUPRC) and the area under the receiver operating curve (AUROC) under the different settings
Summary
Large language models (LLMs) have been shown to have significant potential in few-shot learning across various fields, even with minimal training data. Obtaining cell lines from these tissues can be physically difficult and expensive, which limits the number of training data available for drug pair synergy prediction This can make it challenging to train machine learning models that rely on large datasets. Our experimental results demonstrate that our LLM-based few-shot prediction model achieved significant accuracy even in zero-shot setting (i.e., no training data) and outperformed strong tabular prediction models in most cases This remarkable few-shot prediction performance in one of the most challenging biological prediction tasks has a critical and timely implication to a broad community of biomedicine because it shows a strong promise in the “generalist” biomedical artificial intelligence[1]
Sign-in/Register to view highlights & summary 
Published Version (
Free)
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
