Post by ummefatihaayat22 on Feb 13, 2024 4:49:38 GMT -5
Detection and relationship of genes and rare diseases in texts Who has never entered a Christmas card contest? For many of us, those art contests in which all the school students invested hours trying to draw the best Christmas picture are far away. However, currently, it is possible for even a child to draw a Christmas like the one in the following image in just minutes thanks to artificial intelligence , specifically a model called Stable Diffusion. Stable Diffusion is a text-to-image model, that is, a model that waits to receive a text called a prompt to generate an image that represents said text . The operation of Stable Diffusion and diffusion models in general to generate images from texts has already been explained in this blog. However, in this post we are going to focus on some less technical, but no less important, questions, such as what type of prompts generate the best images? Can I generate the image I want with an AI?
Given the large amount of text information that is generated in the health sector, applications arise for the extraction and relationship of terms that facilitate the work of reading and obtaining knowledge of medical professionals. At the Institute of Knowledge Engineering (IIC), in collaboration with the Progreso y Salud Foundation, we have recently developed GER, a search platform that allows us to detect and relate rare genes and diseases in biomedical texts.
Through Natural Language Processing (NLP) techniques and with different visualization options , the application allows you to quickly find the relationships between rare diseases and genes of interest and speed up the “reading” of scientific Germany Telemarketing Data articles on the subject. How does GER work? In the GER video demonstration we see how, after performing a search for a rare disease, a series of documents containing the entered term appear , ordered from most to least relevant. If we enter a specific document, we find the rare diseases highlighted in pink and the genes related to said disease in yellow, being able to see at a glance those that appear in the same text and obtaining more structured texts.
In addition, these relationships can be seen in the form of a graph , offering more visual information. Each link between gene and rare disease indicates that they have appeared at least once in the same biomedical article.NLP in genomics A clear potential of NLP in health is seen especially in the field of genomics. In the end, the information is written in some way in papers or databases and it is very difficult to find all the details manually. Thanks to NLP, it can be reached more quickly by training a system that highlights what is relevant for the professional or for each case.
In this way, data analysis and genomics come together to better understand genetic diseases and be able to relate them to other characteristics of patients, so that they can be offered more personalized medicine and treatments . At the IIC, we continue working on the development of systems that identify and relate genomic terms in different medical texts (reports, scientific articles or the clinical history itself). Tips for doing prompt engineering Prompt engineering is the job of finding a text that, when introduced into a diffusion model, generates an image that is as close as possible to the one we have in our imagination . This is often not easy, since it must be remembered that diffusion models in no case have the capacity to reason or imagine, they simply limit themselves to drawing images similar to those they have seen during their training using numerical representations of the text and images. images that completely escape our understanding.