- Antimicrobial resistance poses a major threat to public health.
- Researchers used AI to discover a new class of antibiotics to treat drug-resistant staph infections.
- The study used graph-based searches for chemical substructure options.
For the first time in over 60 years, a new class of antibiotics to treat drug-resistant staph infections has been discovered using artificial intelligence (AI) machine learning; a landmark breakthrough to address the antimicrobial resistance (AMR) crisis.
This important discovery to advance global health care has been made using artificial intelligence machine learning by researchers affiliated with the Massachusetts Institute of Technology (MIT), Harvard University, and the Broad Institute of MIT and Harvard in Cambridge, Massachusetts.
Antimicrobial resistance is a leading cause of death globally, and a public health threat. A projected 10 million will die annually by 2050 due to AMR according to The Review on Antimicrobial Resistance report commissioned by the UK Government.
Globally, 4.95 million deaths were associated with bacterial antimicrobial resistance and 1.27 million died directly from antimicrobial resistance according to a 2019 study published in The Lancet. There are over 2.8 million antimicrobial-resistant infections and 35,000 deaths as a result in the U.S. annually according to the Antibiotic Resistance Threats in the United States, 2019 by the U.S. Centers for Disease Control (CDC).
Antimicrobials are substances that destroy or inhibit growth of microbes such as antivirals, antiparasitics, antibiotics, and antifungals. Antimicrobial resistance (AMR) happens when microbes mutate or adapt making antimicrobials ineffective. This natural process can be accelerated by improper use or overuse, resulting in harmful viruses, bacteria, parasites, and fungi developing resistance to antimicrobial medicine and antibiotics.
For example, the over-prescribing of antibiotics for humans and overuse in animal livestock feed is resulting in drug-resistant bacteria strains. According to the A Review of Antibiotic Use in Food Animals: Perspective, Policy, and Potential report by Landers and et. al., 88% of swine are fed antibiotics such as tetracyclines or tylosin, 42% of beef calves are fed tylosin, and nearly all dairy cows receive post-lactaction prophylactic doses of antibiotics such as penicillins, beta-lactam, or cephalosporins.
According to the 2022 Global Antimicrobial Resistance and Use Surveillance System (GLASS) Global Antimicrobial Resistance and Use Surveillance System (GLASS) report by the World Health Organization (WHO), the median rate in 76 countries is 35% for methicillin-resistant Staphylococcus aureus and over 40% for third-generation cephalosporin-resistant Escherichia coli (E. coli).
Staph, also known as Staphylococcus aureus (S. aureus), is a Gram-positive bacteria that causes a myriad of infections in humans such as skin infections, sepsis, and lethal pneumonia. Methicillin-resistant S. aureus (MRSA) caused over 120,000 deaths worldwide in 2019 according to the Institute for Health Metrics and Evaluation January 2022 report.
“Our study demonstrates that graph neural networks can be better understood and explained using graph-based searches for chemical substructure rationales that recapitulate model predictions,” wrote MIT professor James Collins, Ph.D., and the study co-authors.
The scientists used an AI platform for graph neural networks (GNN) called Chemprop. The AI graph neural networks use information in the molecular bond and atoms of each molecule to form its predictions. Graph neural networks are artificial neural networks that can process graph data structures in order to perform predictions, analysis, and classification tasks.
“An alluring implication of the present study is that deep learning models in drug discovery can be made explainable,” the researchers wrote.
The scientists screened over 39,300 compounds for growth inhibitory activity of methicillin-susceptible strain, S. aureus RN4220 that resulted in 512 active candidate compounds. The screening data was used to train ensembles of AI graph neural networks to predict whether or not a new compound inhibits bacterial growth based on the atoms and bonds of its molecular chemistry.
An important aspect of drug discovery is to eliminate compounds that may damage or are toxic to human cells. Thus, the researchers counter-screened the training database of the over 39,300 compounds to predict cytotoxicity as well. For understanding general cell toxicity as well as liver toxicity, they counter-screened for cytotoxicity in human liver carcinoma cells (HepG2). For insights into in vivo cell toxicity, the researchers counter-screened human lung fibroblast cells (IMR-90) and human primary skeletal muscle cells (HSkMCs).
From the orthogonal models to predict cytotoxicity, 40% of the 512 active antibacterial candidate compounds were cytotoxic, resulting in 306 without any cytotoxicity for the three cell types used for screening.
Next, the scientists retrained ensembles of 20 AI models with each of the complete training databases to have four AI ensembles to predict antibiotic capabilities, and cytotoxicity for the three types (HepG2, HSkMCs, and IMR-90) of cells. These four AI ensembles were fed input data on over 12 million compounds consisting of over 11.2 million from the Mcule purchasable database and over 799,000 from m a Broad Institute database.
After further screening, the researchers had a set of 283 compounds which were put to experimental growth inhibition testing against MRSA in the lab. This led to the discovery of two antibiotic candidate compounds that were then tested in mice.
Reported Collins and et. al.,
“Our approach revealed multiple compounds with antibiotic activity against S. aureus. Of these, we found that one structural class exhibits high selectivity, overcomes resistance, possesses favorable toxicological and chemical properties, and is effective in both the topical and systemic treatment of MRSA in mouse infection models.”Copyright © 2023 Cami Rosso All rights reserved.
Leave a Reply