Press "Enter" to skip to content

Machine learning in action during drug discovery


In drug discovery, we
can use machine learning to detect very subtle changes in cells that are produced by different compounds and thereby find out how compounds work. To run an experiment
using different compounds, oftentimes thousands of compounds, we use a plate containing
thousands of buckets, or wells, each of which
contain thousands of cells. And each of these buckets is essentially a different experiment with a different compound
treatment to each one. As we run an experiment we treat the cells with
different compounds in each of these different buckets and then use a microscope
to image the cells and those images form the input into the machine for learning. So the computer knows what
a normal cell looks like. We basically give the computer a whole bunch of examples of what these normal
untreated cells look like and from there it can
learn what the signature, or fingerprint, of an untreated cell is and define all other
treatments against that one. For example, we may have a compound that is causing the nuclei to divide but the cell itself doesn’t. So you can see these
sort of multi-lobed blobs within the cell and that gets highlighted using a nuclear stain. In another example, you may have a cell in which
holes appear in the cytoplasm, and so those appear as
like very distinct spots or openings within the cell,
the cell structure itself. And both of those are different than the normal untreated cells. So, from the machine learning process, we’ve basically learned
that different compounds certainly cause the cells
to assume different forms and all it needs to do is basically look for something that’s different than the normal
cells, in whatever fashion. And this machine can do that
much quicker, much faster, and can look for much more subtle features than the human eye can probably see.

Be First to Comment

Leave a Reply

Your email address will not be published. Required fields are marked *