Slashdot Mirror
New Antibody Attacks 99% of HIV Strains (bbc.com)
An anonymous reader quotes a report from BBC: Scientists have engineered an antibody that attacks 99% of HIV strains and can prevent infection in primates. It is built to attack three critical parts of the virus -- making it harder for HIV to resist its effects. The work is a collaboration between the US National Institutes of Health and the pharmaceutical company Sanofi. Our bodies struggle to fight HIV because of the virus' incredible ability to mutate and change its appearance. These varieties of HIV -- or strains -- in a single patient are comparable to those of influenza during a worldwide flu season. So the immune system finds itself in a fight against an insurmountable number of strains of HIV. But after years of infection, a small number of patients develop powerful weapons called "broadly neutralizing antibodies" that attack something fundamental to HIV and can kill large swathes of HIV strains. Researchers have been trying to use broadly neutralizing antibodies as a way to treat HIV, or prevent infection in the first place. The study, published in the journal Science, combines three such antibodies into an even more powerful "tri-specific antibody." The experiments conducted on 24 monkeys showed none of those given the tri-specific antibody developed an infection when they were later injected with the virus. "We're getting 99% coverage, and getting coverage at very low concentrations of the antibody," said Dr Gary Nabel, the chief scientific officer at Sanofi and one of the report authors.
Welcome to core Statistics.
This would have to be a randomized controlled experiment, and the confidence interval being tested would be 99%. What this means in frequentist statistical terms is that if you had 100 test subjects, and out of those you would expect for whatever reason one of those would somehow turn up positive, then you would still be within your 99% confidence interval. More formally stated, the true population mean is somewhere greater than the 2.5th percentile and less than the 99.5th percentile of the the distribution of the values in your samples.
So, because they are working with statistical sampling methods, they never say that they are 100% confident.