This project focuses on the responsible use of AI in predicting protein-ligand binding energies, which is crucial for drug discovery and toxicology. The Nordic and Baltic countries have strong pharma and biotech sectors that rely heavily on computational approaches to design new drugs and medical devices. Molecular recognition, the interaction between biological macromolecules and smaller molecules, is key for designing new medicines and assessing the toxicological effects of pollutants. Measuring the protein-ligand binding energy is the most fundamental way to quantify molecular recognition.
AI, particularly machine learning (ML) and deep learning (DL), has been widely used to predict protein-ligand binding energies. However, the rapid development of AI methods and the lack of standardised tests make it challenging to evaluate their accuracy and reliability. Additionally, the databases for molecular recognition are often lacking, leading to concerns. The business sector has recently warned about the growing gap between perceived progress and real-world impact of AI in this field. A discussion of the responsible use of AI in the calculation of protein-ligand binding energies is lacking almost completely. The project highlights the need for responsible use of AI in protein-ligand affinity prediction, focusing on several key aspects:
- Accuracy: AI models must be compared to experimental data, considering the limitations of these experiments which are often ignored.
- Real Cost: The development, testing, and deployment costs of AI models must be transparently reported.
- Data Availability: There is a need for high-quality protein-ligand affinity databases to support AI algorithms.
- Transferability: AI models should be applicable to various datasets, even those not included in the training and test sets.
- Lack of Insight: AI models often function as black boxes, making it difficult to understand the underlying chemical, biological, and physical processes.
Ultimately, it aims to develop guidelines for the responsible use of AI in predicting protein-ligand binding energies. By establishing these guidelines, the proposal seeks to ensure that AI methods are used responsibly in drug discovery and toxicology, ultimately benefiting scientific research and public health.
The multinational, interdisciplinary consortium that will work on this involves scientists with expertise in biology, computational chemistry, computer science, cancer research and toxicology.
