Photo credit: www.sciencedaily.com
The rise of drug-resistant infections, particularly from severe pathogens like tuberculosis and Staphylococcus aureus, poses an increasing threat to global health. These resistant strains complicate treatment, often necessitating more expensive and potentially toxic medications, which can lead to extended hospital stays and heightened rates of mortality. As noted by the World Health Organization, around 450,000 individuals were diagnosed with multidrug-resistant tuberculosis in 2021, with the success rate for treatment plummeting to merely 57%.
Researchers at Tulane University have created an innovative artificial intelligence-based system that enhances the detection of genetic markers associated with antibiotic resistance in Mycobacterium tuberculosis and Staphylococcus aureus. This advancement could significantly expedite and improve treatment protocols.
The findings from a study published in Nature Communications unveil a novel Group Association Model (GAM) that employs machine learning techniques to pinpoint genetic mutations associated with drug resistance. Unlike conventional methods, which may erroneously correlate unrelated mutations with resistance, GAM operates without presupposing prior knowledge of resistance mechanisms, thereby enhancing its capacity to identify novel genetic variations.
Existing resistance detection methods, utilized by organizations such as the WHO, often suffer from timeliness issues—like culture-based tests—or may overlook uncommon mutations as seen in some DNA assessments. The Tulane model tackles these limitations by analyzing comprehensive genome sequences and comparing diverse groups of bacterial strains showcasing varying resistance profiles to identify genetic alterations that reliably correlate with resistance to specific antibiotics.
“Consider it as leveraging the complete genetic profile of the bacteria to unveil its immunity against certain antibiotics,” stated Dr. Tony Hu, the senior author and the Weatherhead Presidential Chair in Biotechnology Innovation, along with his role directing the Tulane Center for Cellular & Molecular Diagnostics. “We’re effectively teaching a computer to discern resistance patterns autonomously.”
Throughout the study, the researchers applied GAM to over 7,000 strains of Mtb and nearly 4,000 strains of S. aureus, successfully pinpointing critical mutations linked to resistance. Notably, GAM matched or surpassed the accuracy of the WHO’s resistance database while significantly minimizing false positives, which can result in misdirected treatment.
“Existing genetic tests sometimes misidentify bacteria as resistant, which can compromise patient management,” commented lead author Julian Saliba, a graduate student at the Tulane University Center for Cellular and Molecular Diagnostics. “Our approach offers a more accurate representation of which mutations genuinely confer resistance, helping to lessen misdiagnoses and avoid unnecessary treatment modifications.”
When combined with machine learning, this model demonstrated improved predictive power for resistance, even when working with limited or incomplete data. Validation studies utilizing clinical samples from China indicated that the machine-learning-enhanced model had superior performance compared to WHO-based methods in predicting resistance to essential first-line antibiotics.
This advancement is particularly crucial, as early detection of resistance enables healthcare providers to tailor effective treatment strategies before an infection has the chance to escalate.
The flexibility of the model to identify resistance without relying on predefined expert guidelines suggests its potential applicability to various bacterial strains, as well as in agricultural settings where antibiotic resistance remains a pressing issue in crop management.
“Staying ahead of the continually evolving landscape of drug-resistant infections is imperative,” emphasized Saliba. “This innovative tool could be a significant asset in that endeavor.”
Source
www.sciencedaily.com