Up until recently, pharma-companies begin drug development with a hypothesis that a particular protein target and a known biological mechanism can be manipulated to produce medicine. Most often, they place a risky wager on whether the protein targets are indeed responsible for the diseases that they want to cure. However, a new generation of start-ups is using artificial intelligence (AI) to analyse vast databases of clinical and biological data without resorting to preconceived notions. Instead, the AI itself will help with hypothesis generation.
Human genetic information, primarily from population-scale studies known as genome-wide association studies (GWAS) that compare the genetic characteristics of patient cohorts to those of healthy controls, has sparked several drug programmes over the past 20 years. Additionally, the growth of sizable research biobanks and national public-private collaborations like Genomics England, which has gathered phenotypic and genomic data from over 150,000 individuals, has provided medicine firms with a wealth of material to work with.
This quantity of data makes it considerably simpler to identify uncommon gene mutations that have a significant impact on health and illness. However, when these databases grow and include other -omics data such as transcriptomic, proteomic, and metabolomic, they are harder to evaluate. This is where AI may be a useful tool, especially when looking for signals in the data that might not be immediately visible. This is because AI and ML are pretty good at looking across a broad swath of variables at really subtle nonlinear signals.
AI may be used at several phases of the study, including the beginning, when it is sifting through the full haystack of biological data to find a little fragment of useful information. A researcher can merely ask for a list of the most notable correlations between this genotype and any trait. Alternately, it might concentrate on certain disease phenotypes and more limited subsets of genes and pathways to provide biological justifications for particular illnesses. The AI’s output is often simply the first step toward target identification, even with the most potent algorithms. In the larger framework, closing the loop in drug discovery is more important. You may develop this hypothesis using neural networks, and then send the target candidate to experimenters for testing. The results of those tests can then be used to guide model learning once more.