Artificial intelligence’s transformative impact on genetic science re-emerged via Alphagenome, which is an AI tool developed by OpenAI. Developed to decode the complex structure of the genome and shed light on the non-coding regions of DNA, the model stands out for its capacity to predict the molecular effects of regulatory variants. AlphaGenome is not just an AI tool; it also provides a comprehensive analysis platform that is capable of bringing significant changes to genetic research.
The model shifts the focus to the silent parts of the DNA and reveals their crucial roles in gene regulation. The human genome is not fully made of protein-coding sequences. 98% of our genome is non-coding regions, and most of these regions play roles in gene regulation. However, to this day, difficulties have been encountered in deciphering the biological functions of non-protein-coding sequences, both experimentally and computationally. To solve these problems, AlphaGenome processes base pairs up to 1 million pairs long; this way, it determines the molecular regulatory activities of various cell and tissue types. Additionally, it can calculate the molecular effects of mutations in seconds. Thus, it enables us to understand the result of small changes that happen on DNA.
AlphaGenome uses a neural network that has deep learning as its core, which facilitates molecular mapping. This structure is able to establish relationships between different parts of the sequence, and in this way, it can model the detailed structure of gene regulation better. The model makes use of experimental data from prominent databases such as ENCODE, GTEx, FANTOM5, and 4D Nucleome. This data includes thousands of molecular properties like RNA production, accessibility of chromatin, and gene expression in human and mouse tissues.
Evaluating the effects of genetic variants has long posed a major challenge for geneticists. This new tool is able to calculate the effects of variants in seconds. AlphaGenome accelerates this process by comparing the molecular properties of mutated DNA sequences with reference sequences, allowing it to estimate the impact of variants on regulatory functions. This provides researchers with significant advantages in identifying the causes of diseases and in developing targeted therapeutic approaches in genomic medicine. The model’s contribution is particularly valuable in complex biological processes such as RNA splicing. AlphaGenome can predict which regions of RNA are retained or excised directly from the sequence information. This is of great importance for understanding the mechanisms underlying splicing errors that lead to genetic diseases.
The model outperformed other genomic prediction systems. In 22 of 24 independent tests on separate DNA sequences, AlphaGenome outperformed other models. It is placed first in 24 out of 26 tests that evaluated variant effect prediction. Surprisingly, this level of performance was achieved even when compared to models created specifically for single tasks, while AlphaGenome can manage all of the jobs at once.
The tool can be used in various fields, including mapping the roles of gene regulatory factors, demonstrating uncommon disease-causing mutations, and in synthetic biology, creating DNA sequences that become active only in one type of cell. For instance, a study on T-cell acute lymphoblastic leukemia demonstrates that Alphagenome is not only a prediction tool but also a hypothesis generator, by hypothesizing that specific variations may indirectly activate the TAL1 gene, thereby rediscovering a known disease mechanism. This indicates the model’s potential as a hypothesis generator for biological discovery, not only as a prediction tool.
Despite AlphaGenome’s various novel features, there are still a number of issues that need to be eliminated. The major issue is modeling the interactions between regulatory regions on DNA that are placed far from each other, and it is hard to accomplish because of difficulties in both data diversity and processing power. Another challenge is in personalized medicine applications, where the model is not yet optimized for direct predictions on individual genomes. However, AlphaGenome’s capabilities give a strong base for resolving these difficulties in the following advancements.
The AlphaGenome API currently establishes the model for non-commercial usage, and permissions are limited to research purposes. There are still unfinished validation procedures for clinical applications. However, OpenAI aims to collaborate with the science community to build the system further and improve it frequently with the help of feedback from the research community.
AlphaGenome’s future improvements, which may include genomic data from various species, may contribute to population genetics, evolutionary biology, and even ecosystem biology.
Author: Büşra Yaşar
Editor: Şimal Yıldız
Reference: , , N., et al. AlphaGenome: Advancing regulatory variant effect prediction with a unified DNA sequence model. Biorxiv. (2025). https://doi.org/10.1101/2025.06.25.661532
-Bioinfocodes Scientific News Service-
News articles prepared by our team members, reviewing and compiling scientific research published in journals with and impact factor greater than 20 (click here for the list).