Computer based candidate identification

Big data and AI can also be used for in silico (computer-based) identification of drug repurposing candidates. This approach can use algorithms to mine available literature databases and extract information relevant for a particular drug or disease (e.g. disease mechanisms, genetic mutations, small molecules that could affect the disease or a list of authors or institutions looking at the condition). AI can even be used to link published studies, infer connections between them and make predictions about drugs/drug combinations.

Multi-omics approaches (studying the genes and proteins in an individual) can also be used to find repurposing opportunities. For example, if a gene is overexpressed in a disease, a drug that causes its down-regulation could offer a potential therapy.

A series of four bar graphs to illustrate the expression matching approach used by many artificial intelligence approaches to identify repurposing opportunities. Each graph shows a series of different genes as bars on the x-axis (horizontal), and the vertical axis (y-axis) shows the level of gene expression. Bars above the lines represent genes that are over expressed (there is too much made compared to normal), bars below are genes that are under expressed (there is less produced than needed). The first graph shows a series of genes that are over expressed and under expressed for a rare condition. Next to this is a graph showing that a drug can produce an opposite expression patten in a normal cell – the drug causes the genes that are over expressed in the disease to be under expressed by a similar amount. When these are added together – the diseased cell is given the drug – the equal but opposite signals combine, to the final graph, which shows most of the bars are completely gone. This means the cell is now behaving normally, and the disease signature, and hopefully the disease, is removed.

Using this information, repurposing opportunities can be predicted, and new chemical entities modelled and developed using computers.

A successful example of a drug repurposing partnership using AI was exemplified by a collaboration between Healx and the Fragile X Association of America. Using their AI technology to establish which drug expression profiles would counterbalance the aberrant gene expression profiles for the disease, Healx were able to generate 18 candidate drug combinations which had potential to treat fragile X syndrome. Using their fragile X mouse models, the Fragile X Association of America were then able to screen these drug combinations with relatively high throughput and found that 9 of the 18 combinations reversed all four fragile X phenotypes in mice. These candidates will be taken through to clinical trials, which are planned to commence in 2020.