Researchers from the United States and China have developed a new AI model that could help overcome obstacles in drug development and discovery. The new AI model, named ActFound, outperforms competitor models and has proved to be a more cost-effective alternative to traditional methods for bioactivity prediction.
Bioactivity includes a compound's interactions with biological targets, effects on systems, and therapeutic outcomes. Predicting bioactivity is essential for identifying potential useful compounds from a large pool of candidates, saving time and costs in drug development and experiments.
The primary challenges in applying AI and machine learning (ML) to bioactivity prediction stem from insufficient data labeling and the inconsistencies between assays, which are the tests that evaluate the activity or potency of the drugs.
The team of researchers from the University of Washington, Peking University, and the AI tech firm INF Technology Shanghai detailed the new model in a paper published in Nature Machine Intelligence.
Not only does the new model outperform other competing AI models, but also functions as a free-energy perturbation (FEP), a well-established traditional computational method used in drug discovery.
The researchers emphasized that FEP calculations are computationally demanding as they “require extensive computational resources that are often not affordable for large-scale applications". While the FEP method delivers excellent accuracy, such methods often require data that is elusive and needs expensive equipment and extensive laboratory procedures to obtain.
Using ActFound, the researchers were able to use fewer data points while maintaining high accuracy. This makes it a significantly less expensive alternative to FEP.
“Our promising results indicate that ActFound could be an effective bioactivity foundation model for various types of activities,” said Wang Sheng, corresponding author and assistant professor at the University of Washington.
China has been keen to invest in research and development for its booming pharmaceutical industry. Several key players in the industry are harnessing the power of AI to cut development time. AI enables them to evaluate the bioactivity of compounds more efficiently and cost-effectively than ever before.
ML methods often struggle with bioactivity prediction due to the limited number of compounds tested in each assay. The existing assay data can also suffer from inconsistency, making it challenging for models to generate from one assay to another. Foundation models that are pre-trained data on large and diverse data sets can overcome this challenge as they can make predictions on new and unlabeled data more effectively.
Using this approach, ActFound was trained on 35,644 assays from a trusted and well-known chemical database and 1.6 million experimentally measured bioactivities. This helps improve the model's accuracy, generalizability, and ability to capture complex patterns in bioactivity predictions.
The researchers used two machine learning methods: meta-learning and pairwise learning.
The Meta-learning method enables the model to make predictions even with limited data by drawing on knowledge from a large number of assays. This is vital for drug discovery, where generating extensive bioactivity data can be costly and time-consuming.
The role of the Pairwise model is to help generalize the model by comparing compounds relative to each other instead of predicting exact values. The researchers shared that it was their intuition that compounds from different assays would be comparable, which led them to take the novel approach of combining pair meta-learning and pairwise learning methods in a single model.
ActFound was evaluated using six real-world bioactivity datasets and proved to be more effective than nine other models within the same domain and across different domains. This highlights its ability to not only predict bioactivity for data it is trained on but also perform well with new types of data.
The model was also tested in a case study to predict the bioactivity of cancer drugs, and the researchers reported that it performed better than other models.
The development of ActFound marks a significant step forward in using advanced technologies for drug development and discovery. AI and ML have been at the forefront of drug discovery research and development. They are paving the way for new breakthroughs and accelerating the discovery process.
Related Items
Google DeepMind’s New AlphaFold Model Poised to Revolutionize Drug Discovery
New AI Technique Helps Find Alzheimer’s Drug Targets
MIT Researchers Leverage AI to Identify Antibiotic That Can Kill Drug-Resistant Bacteria
