LucaVirus - Alibaba's open-source fundamental virus model

What is LucaVirus?

LucaVirus is a unified nucleic acid-protein language model developed by Alibaba Cloud LucaGroup specifically for viruses. Trained on 25.4 billion nucleotide and amino acid tags, it covers almost all known viruses. The model can learn biologically meaningful representations such as the relationships between nucleotide and amino acid sequences. Downstream models developed based on this model can address key challenges in virology, such as identifying viruses in the "dark matter" of the genome, characterizing unknown protease activities, predicting viral evolutionary capabilities, and discovering antibody drugs against new viruses, demonstrating outstanding performance in these tasks. Its protein embeddings can distinguish protein families at high resolution, exhibit strong correlation between embedding distance and genetic distance, and provide rich evolutionary information. It also demonstrates superior performance in antibody-antigen binding prediction, with accuracy and other metrics exceeding existing models and structure-based prediction methods.

Main functions of LucaVirus

• virus discoveryIt can identify viruses hidden in the "dark matter" of the genome, helping scientists discover new viral sequences in complex genomic data and expand their understanding of virus species.
• Functional predictionIt can characterize the enzymatic activity of unknown proteins, predict their potential biochemical functions by analyzing protein sequences, and provide clues for understanding the pathogenic mechanisms of viruses and developing antiviral drugs.
• Evolutionary analysisPredicting the evolutionary capacity of viruses and modeling the evolutionary information of viral sequences helps researchers understand the mutation trends and evolutionary paths of viruses, which is crucial for virus monitoring and control in the public health field.
• Drug discoveryThe goal is to discover antibody drugs targeting emerging viruses, use models to predict the binding potential between viral antigens and antibodies, accelerate the development of antibody drugs, and improve the ability to respond to emerging infectious diseases.

The technical principles of LucaVirus

• Multimodal data fusionBy combining nucleotide and amino acid sequence data, a unified nucleic acid-protein language model is constructed to learn the complex relationship between the two.
• Large-scale data trainingThe model is trained on 25.4 billion nucleotide and amino acid tags, covering almost all known viruses, ensuring broad generalization ability and understanding of viral diversity.
• Evolutionary Information ModelingBy embedding learning, the evolutionary information of viral sequences is incorporated into the model, enabling the model to capture the evolutionary divergence and homology of viruses, thus providing support for evolutionary analysis.
• Interpretable EmbeddingThe generated embeddings can distinguish protein families at high resolution and correlate them with genetic distance, providing interpretable biological representations for virology research.
• Downstream task adaptationDevelop specialized downstream models to optimize for tasks such as virus discovery, functional prediction, evolutionary analysis, and drug discovery, thereby improving the performance of these models in practical applications.

LucaVirus project address

• Github repository: https://github.com/LucaOne/LucaVirus
• HuggingFace model library: https://huggingface.co/collections/LucaGroup/lucavirus-689d9382d0cc09780f380958

Application scenarios of LucaVirus

• Public health surveillanceBy rapidly identifying emerging viruses and monitoring their evolutionary trends, it provides early warnings to public health departments, helps develop effective prevention and control strategies, and reduces the risk of virus transmission and outbreaks.
• Disease diagnosisIt assists medical personnel in more accurately diagnosing viral infectious diseases, especially for diseases with similar symptoms but caused by different viruses, thereby improving the accuracy and efficiency of diagnosis.
• Vaccine developmentIt provides crucial information for vaccine development, such as predicting antigenic changes in the virus, helping to design more effective vaccines, improving the vaccine's adaptability to viral mutations, and enhancing the vaccine's protective effect.
• Drug developmentAccelerate the research and development process of antiviral drugs by predicting the function of viral proteins and drug targets, providing a theoretical basis for new drug design and reducing research and development costs and time.
• Biosecurity defenseIn the field of biosafety, it is used to detect and identify potential biological threats, such as the emergence of new viruses, providing technical support for national and regional biosafety and safeguarding public health and social stability.