One of the greatest challenges in treating tuberculosis the top infectious killer worldwide, according to the World Health Organization (WHO) is the bacterium’s ability to shape shift rapidly and become resistant to multiple drugs. New tests that can quickly and reliably detect resistance to the most commonly use drugs before a patient begins treatment are urgently need to improve outcomes and help curb the spread of the infection.
Now, Harvard College apply math student Michael Chen 20; hence working with biomedical researchers at Harvard Medical School’s Blavatnik Institute, has design a computer program that sets the stage for the development of such tests. The program, describe April 29 in EBioMedicine, can accurately predict a TB strain’s resistance to 10 first; also second-line drugs in a tenth of a second and with greater precision than similar models.
Neural network model
Our goal was to develop a neural network model; which is a type of machine learning that loosely resembles how connections between neurons are form in the brain,” said Chen, who start writing the program as a first-year and is now a junior pursuing a degree in apply math with a secondary degree in computer science. The study’s first author, he plans to go to medical school.
The wide-and-deep model is a decision making tool that combines all of the data; hence with prior biological knowledge that resistance is cause both by large individual mutations and the interactions between many different mutations,” said study co-senior author Andy Beam; hence faculty member in biomedical informatics at HMS and a visiting lecturer at the Harvard T.H. Chan School of Public Health.
The models were train on 3,601 TB strains resistant to first and second line drugs; which including 1,228 multidrug-resistant strains. The data set include results from drug-sensitivity testing; so allowing the model to assess the link between the presence of certain mutations and drug sensitivity. The training sets were generate with data from various countries curate by researchers at the Critical Path Institute; so with support from the Bill & Melinda Gates Foundation.
To test their performance in a realistic setting, the models were challenge to predict resistance; so in a set of 792 fully sequence TB genomes they never “see” before. This create a high fidelity testing scenario that eliminate the chance that the models would encounter familiar strains they had “studied” during training.
Network model predict resistance
The wide and deep neural network model predict resistance to first-line drugs with 94% accuracy; hence on average, and 90% accuracy to second-line drugs, on average. The statistical model predict resistance with 94% and 88% accuracy, respectively. The wide-and-deep neural network model showed slightly greater accuracy; so in predicting resistance to three specific drugs.
Both models are capable of predicting resistance to first and second line therapies in a tenth of a second; hence underscoring the increasingly important role that artificial intelligence and big data will play in the rapid detection and choice of therapy for the disease.
Our model highlights the role of artificial intelligence in the case of TB; but its importance goes well beyond TB,” Beam said. AI will help guide clinical decision-making by rapidly synthesizing large amounts of data to help clinicians make the most informed decision in many scenarios and for many other diseases.