Hey, thanks for putting this together! A few high level questions my group and I had regarding the challenge?
- What will happen downstream of platform for classifying active antibodies? Have the 48 antibodies in the training set been tested in a clinical setting? We're curious how making a model for the open task will be used for discovery. - How is the data for the open task of antibodies against bacterial infection specific to covid-19? Is there something that makes the 'secondary infection problem' easier, or a subset of the general problem of identifying antibiotics that treat drug-resistant bacteria? - Is it known how many people die because of secondary infection, as opposed to just with it? From my understanding of the literature it seems many who don't survive do indeed have secondary infection; however how often is it the cause among cytokine storm/inflammatory response/etc?
Hi Ron,
Glad to answer your questions.
1. We are not classifying active antibodies, but small molecules. The description of the dataset provides information on how they were tested on cell lines. This is a standard way of screening the data for their effectiveness. You cannot put a drug in humans before it is validated as safe.
2. We summarize this information in Lancet article posted in the blog. In summary, 95% of patients hospitalized in Wuhan (described in that article) were given antibiotics. Unfortunately, they didn't help to many patients as the bacteria was drug resistant. There are many possible types of bacteria that can cause pneumonia. Pseudomonas is just one type, which is hard to treat.
3. We are not medical professionals, so we cannot answer this question. We will reach out to those who know more, and update the post if something new becomes known.