Deadly viruses that infect humans often originate from other animals. COVID-19 is simply one example. From aggravating poverty traps to contributing to rising mental health issues, the implications of such viruses are wide-ranging and long-lasting. Thus, using technology to address this problem would be highly valuable.
At the University of Glasgow, researchers have employed machine learning that traverses a virus’s genetic makeup, and then predicts the probability of it spreading to humans. The in-depth paper is titled, “Identifying and prioritizing potential human-infecting viruses from their genome sequences.”
To do this, they used training data containing around 860 viruses from 35 families, which were individually compared against genomes that typically infect humans, and given a probability accordingly. After experimenting with different models, they used the most effective one to analyse the DNA patterns of 645 more viruses. Further tests were done to predict the ‘zoonotic potential’ of known coronavirus species.
Of course, the ML isn’t perfect. There will false positives and negatives, so should be confirmed by actual lab tests. Meanwhile, there are other factors that come into play apart from whether they can infect humans or not: the surrounding conditions needed for survival, transmission rate between humans, and so on so forth. The initial model is quite promising already though, given that it helps filter out threatening viruses in a low-cost, fast manner. Moreover, with more genome data and fine-tuning, the Ml will inevitably become smarter over time.
If successful, this means scientists can create vaccines before disaster strikes, improving distribution capabilities and efficacy rates.
Written by Amanda Y