Large Language Models in Drug Discovery and Development
Developing a new drug is a long and expensive process, which often takes 10-15 years and costs more than US$ 2 billion (Berdigaliyev & Aljofan, 2020). It contains three main steps: understanding the disease and identifying a treatment target, designing an appropriate drug, and testing it in clinical trials. Each step is composed, time-consuming, and resource-intensive, delaying the arrival of new treatments.
Artificial intelligence (AI), especially large language model (LLM), provides promising methods for speeding up this process. LLMs can understand scientific texts, support disease research, help with drug design and clinical testing, and help handle data. Recent models have been used for tasks such as identifying the goals of the drug, automating chemical experiments, and analyzing of clinical knowledge. With further progress, LLM drugs can play an important role in all stages of development, which can help improve efficiency and results.
In the discovery of the drug, scientific languages as SMILES (used to represent molecules) and FIXED (used for protein and genetic sequences), are crucial to coding biological information. To interpret these formats effectively, two main types of language models are used: special models and general objective models.
2.1 Specialized Large Language Model
Specialized LLM is trained on scientific data and is designed to handle specific languages used in biology and chemistry. These models can extract meaningful patterns from raw research.
In disease research, they help by analyzing gene expression data and DNA sequences and revealing insights such as genetic markers, regulatory elements, and genetic networks. Some protein-based models can predict the structure or function of the protein by analyzing amino acid sequences and improving understanding of the disease system.
In the discovery of the drug, these models help predict chemical reactions, plan synthesis, and suggest new molecules with desired properties. They also support the first safety screening by predicting absorption, metabolism, and toxicity (ADMET).
Usually, these models work to take defined inputs as a protein sequence and a molecule SMILES code and return a prediction, such as binding power or the possibility of interaction.
2.2 General purpose Large language model
General Purpose LLM is trained on a wide range of lessons, including scientific literature, so that they can understand complex biomedical materials and apply it to different fields.
They can review the huge versions of the published data, summarize the findings, create a relationship between genes and diseases, and support the target identity. These models also help explain technical words in single languages, making it easy to understand and communicate scientific material.
Comments
0 comment