A team of researchers led by Yuxin Yang and colleagues has developed an innovative deep learning framework, named LISA-CPI, that combines molecular imaging and protein structural representations to identify potential drug candidates for pain treatment.
Chronic pain is a major global health issue, and traditional pain management options, like opioids, are associated with severe side effects such as addiction. To address this challenge, Yang et al. introduced a novel method leveraging artificial intelligence (AI) to accelerate the discovery of non-opioid painkillers, targeting specific G-protein-coupled receptors (GPCRs) involved in pain signaling pathways.
LISA-CPI is unique in that it integrates molecular images of drug-like compounds and 3D structural representations of proteins from the advanced AlphaFold2’s Evoformer algorithm. This approach allows for highly accurate predictions of compound-protein interactions (CPIs), with the model trained on over 10 million unlabeled molecules and evaluated on 104,969 ligands interacting with 33 pain-related GPCRs. Compared to existing models, LISA-CPI showed a remarkable 20% improvement in predictive accuracy, highlighting its potential to revolutionize the field of computational drug discovery.
One of the major breakthroughs of this study is LISA-CPI’s ability to identify repurposable drugs, meaning drugs that were originally developed for other conditions but may also be effective for pain management. Among the compounds identified were methylergometrine and gut metabolites like citicoline, which showed promising interactions with pain-related GPCRs. These findings open new avenues for pain therapy, especially focusing on non-opioid targets, which could significantly reduce the risk of addiction and other adverse effects associated with current pain medications.
The deep learning model was rigorously tested on a wide range of data from the ChEMBL and GLASS databases, and it consistently outperformed other machine learning methods, such as ImageMol and CHEM-BERT. LISA-CPI's superior accuracy stems from its dual focus on chemical awareness through ligand imaging and detailed protein structure understanding through 3D protein residue pair representations. This combination makes it a powerful tool for exploring drug-protein interactions that are critical in pain perception and other complex diseases.
In addition to drug repurposing, LISA-CPI also explored the potential of gut microbiota-derived metabolites in pain treatment. Gut health has been increasingly linked to various chronic conditions, including pain. Using LISA-CPI, the team discovered that certain metabolites, such as citicoline and NAD, produced by gut bacteria like Bacteroides, may have therapeutic potential in modulating pain by targeting GPCRs. This insight adds a new layer of complexity to pain treatment, suggesting that manipulating the gut microbiome could become a novel approach to pain management.
This study marks a significant advancement in AI-powered drug discovery, particularly for its application in pain treatment. The use of sophisticated models like LISA-CPI not only improves the accuracy of predictions but also accelerates the process of identifying new therapeutic options, making it a valuable asset in the search for more effective and safer pain relief methods. The full study, including all data and predictions, is available in the journal *Cell Reports Methods* and can be accessed online through the following link (clearnet).
If you're interested in such publications, please react and leave comments. This will be a sign for me to continue.