IBM has developed a generative AI system to speed up the design of broad-spectrum and low-toxic antimicrobials. This approach is targeted at finding newer ways for tackling the menace of antibiotic resistance.
Antibiotic resistance is one of the biggest threats to global health, food security, and development today. Drug-resistant diseases claim 700,000 lives each year globally, and the number is expected to rise to 10 million deaths per year by 2050. Patients who are susceptible to drug-resistant pathogens are also more vulnerable to a viral illness, such as influenza, severe acute respiratory syndrome and COVID-19.
IBM Research AI has focussed its efforts towards Accelerated Discovery by leveraging artificial intelligence-based models for molecular design for antibiotics. The article “Accelerating antimicrobial discovery with controllable deep generative models and molecular dynamics,” published in Nature Biomedical Engineering, proposes Antimicrobial Peptides (AMP) as drug candidates for tackling antibiotic resistance.
“In this Article,… we propose a computational framework for the targeted design and screening of molecules, which combines attribute-controlled deep generative models and physics-driven simulations. For a targeted generation, we propose Conditional Latent (attribute) Space Sampling (CLaSS), which leverages guidance from attribute classifier(s) trained on the latent space of the system of interest and uses a rejection sampling scheme for generating molecules with the desired attributes,” states the article.
IBM’s research has suggested natural AMPs as potential next-generation antimicrobial agents “owing to their exceptional structural and functional variety, promising activity and low tendency to induce resistance.”
AI methods, such as statistical learning and optimisation-based approaches, have shown promise in designing small molecules and macromolecules, including AMPs. “A conventional approach is to build a predictive model that estimates the properties of a given molecule, which is then used for candidate screening,” as per the article.
“Deep-learning-based architectures, such as neural language models as well as deep generative neural networks, have emerged as a popular choice. Probabilistic autoencoders, a powerful class of deep generative models that learn a bidirectional mapping of the input molecules (and their attributes) to continuous latent space, have been used for this design task,” the article adds.
The results have been encouraging. “Thus, the present strategy provides an efficient de novo approach for discovering new, broad-spectrum and low-toxic antimicrobials with therapeutic potential at 10% success rate and at a rapid (48 d) pace.”
