Using AI to explore the distant family trees of proteins
The innovative AI-powered tool, Dense Homolog Retriever, will pave the way for faster drug discovery and enhanced disease research.
Scientists at CUHK have created an innovative AI-powered tool that makes it easier to find connections between proteins, even when those connections are hidden by millions of years of evolution. This breakthrough could speed up drug discovery, improve disease research and help in the fight against antibiotic resistance.
Proteins are the building blocks of life, playing a role in nearly every biological function. Understanding the evolutionary relationships between proteins is crucial for predicting their functions and designing new drugs and clinical therapies. To do this, biologists seek out proteins that share a common evolutionary ancestor and analyse their similarities and differences. However, identifying such relationships is challenging, especially when proteins have evolved significantly over time.
Rapid mapping through machine learning
Until now, scientists have looked for homologs using traditional methods that check for similarities in genetic sequences. However, these methods are slow, time-consuming and often miss connections between proteins that have changed significantly. This makes it harder to map out networks of proteins, which are crucial for understanding diseases and creating treatments.
To solve this problem, Prof. Yu Li and his team at CUHK’s Faculty of Engineering developed a new AI-powered method called Dense Homolog Retriever (DHR). Instead of simply comparing genetic sequences, DHR uses artificial intelligence to encode protein sequences into numerical representations, allowing it to detect similarities more quickly and accurately. By applying machine learning techniques, the AI system can recognise subtle relationships between proteins that older methods might miss.
‘DHR is particularly adept at discovering distant homologs, which are often missed by traditional methods,’ said Prof. Li. ‘It’s not only faster but also more precise, making it ideal for exploring protein sequence diversity across large databases.’
DHR was originally designed to improve AlphaFold 2, an AI system used to predict protein structures. By integrating DHR, researchers significantly sped up the process of finding related proteins, removing a major roadblock in biological research. The team now plans to release an open-source version of DHR so that more scientists can use it.
A wealth of potential medical applications
DHR’s ability to detect distant homologs has the potential to transform several areas of medicine, including cancer research, the fight against antibiotic resistance and the study of neurodegenerative diseases.
In cancer research, the tool can help identify proteins related to tumour growth and cell death, leading to new treatments that target these processes more precisely. In antibiotics, DHR can detect conserved proteins that are essential for bacterial survival across different strains. This information can help scientists design new antibiotics that remain effective even as bacteria evolve.
Neurodegenerative diseases such as Alzheimer’s and Parkinson’s are associated with the misfolding and accumulation of certain proteins. By identifying homologs of these proteins, DHR could provide new insights into the molecular mechanisms of these conditions and guide the development of therapies to prevent or slow disease progression.
The research has been published in Nature Biotechnology and highlighted in Nature Methods.
Read more: