AI assisted by lasers to help predict antibiotic resistance in patients

When germs such as bacteria and fungi develop the ability to defeat the drugs designed to kill them, antibiotic resistance happens. That is bad news, as the germs continue to grow.

Researchers writing in the journal Nature Medicine note that “infections with antimicrobial-resistant pathogens are associated with substantial morbidity, mortality and healthcare costs.” 

In order to have a positive outcome of an infection, rapid treatment is vital. Yet antimicrobial therapy and dosage need to take into account the pathogens’ drug resistance, and the patient’s age, kidney function, previous medical history and the drugs they are already using at the time they become sick with the pathogen.

It takes up to 72 hours to collect a sample and get a result from a test of the pathogen in a culture to determine the best antibiotic treatments, yet time is of essence when there is a sick patient at hand. This means “for a substantial period of time a patient may be receiving an antimicrobial drug with either too narrow or too broad a spectrum.”

The researchers tried to speed up this diagnostic process by using artificial intelligence. This would allow them to be more specific, picking the right antibiotic at the right dosage for the patient, and thereby reducing the use of broad-spectrum antibiotics. This would also be helpful in “combating the development of antibiotic resistance.”

They made use of data from a publicly available database of mass-spectra profiles of clinical isolates. They developed a “novel machine-learning approach” to predict antimicrobial resistance directly from “matrix-assisted laser desorption/ionisation-time of flight (MALDI-TOF) mass spectra profiles of clinical isolates.”

MALDI uses a laser striking a matrix of small molecules, turning them into gas without fragmenting or decomposing them. “MALDI is appropriate to analyse biomolecules like peptides, lipids, saccharides, or other organic macromolecules.”

“Intelligent computer algorithms search the data for patterns that distinguish resistant bacteria from those that are responsive to antibiotics,” says Caroline Weis, a doctoral student in the biosystems science and engineering department at ETH Zurich and lead author of the study in the journal Nature Medicine.

When doctors can identify significant antibiotic resistance at an early stage, they can arrange antibiotic treatment for the patient quicker – this would help very sick patients.

“The time taken to optimise antibiotic therapy might mean the difference between life and death if an infection is serious,” says Adrian Egli, professor and head of clinical bacteriology at the University Hospital Basel. “A fast, accurate diagnosis is extremely important in those kinds of cases.”

The scientists used 300,000 mass spectra profiles with more than 750,000 antimicrobial resistance phenotypes from four medical institutions in northwestern Switzerland. They then linked hese to the results of corresponding clinical resistance tests, Futurity reports. The result, the research site notes, is a new, publicly available dataset covering about 800 different bacteria and over 40 different antibiotics.

“Our next step was to train artificial intelligence algorithms with this data such that they could learn to detect antibiotic resistance on their own,” Karsten Borgwardt, professor in the biosystems science and engineering department at ETH Zurich in Basel, who led the study with Egli, says.

Researchers analysed how the algorithm’s performance was influenced by the training data to make their predictive model as widely applicable as possible – for example, they trained the model with data from one hospital versus later training the model with combined data from multiple hospitals.

According to the ETH Zurich [public research university] press release, “[w]hile previous studies in this field of research have focused on individual bacterial species or antibiotics, this new study draws on several bacterial types isolated in hospitals as well as a multitude of associated resistance characteristics.”

“Our dataset is the largest to date to combine mass spectrometry data with information on antibiotic resistance,” Borgwardt says. “It’s a great example of how existing clinical data can be used to generate new knowledge.”

The scientists conducted a “retrospective clinical case study of 63 patients [and] found that implementing this approach would have changed the clinical treatment in nine cases, which would have been beneficial in eight cases (89 percent). MALDI-TOF mass spectra-based machine learning may thus be an important new tool for treatment optimization and antibiotic stewardship.”

Before AI and lasers could be used in patient care, the scientists will need to carry out a large-scale clinical trial to confirm their findings and the benefits of the new method in a routine hospital setting. “The planning for such a study is already underway,” Egli says.