Researchers are leveraging artificial intelligence and machine learning to help improve the rate at which drugs move successfully from preclinical to clinical development.
According to some research, the overall success rate of potential drug candidates is relatively low; only a small percentage of drug candidates move successfully out of the preclinical (nonhuman) testing stage, and many more fail during other clinical phases. Researchers note many reasons for these failures, though it is often an efficacy question: it may work great on mice in a lab, but that success doesn’t always translate well to an actual patient in a clinic.
Given how costly it can be to develop a compound and move it through clinical trials, only to have it fail, it’s clear new approaches are needed. Researchers have responded by developing an artificial intelligence methodology at the Center for Precision Computational System Network (PreCSN) that may help predict and choose drug candidates that will be successful in trials.
This methodology, recently described in a Nature Communications article, helps researchers accurately, and with great detail, map a disease and evaluate a potential drug candidate before it even enters preclinical studies. Researchers used their methodology to explore a complex disease, inflammatory bowel disease (IBD) and analyzed extensive information about different gene expressions that could cause IBD, allowing them to map gene variations that were applicable to any IBD patient.
Using this data collected from the program, researchers were able to select a viable drug candidate that could target a common feature of IBD, leaky-gut cells, and used it to conduct a “Phase 0” study, a pre-preclinical investigation of a drug candidate. Researchers tested this drug candidate by collecting cells from biopsies performed on IBD patients and growing them in a dish to imitate actual leaky-gut tissue. Results showed that the cells grown in a dish were responsive to the drug candidate, even though the experiment was performed ex vivo. Their approach further used additional data (existing FDA trials of drugs with similar targets) to make predictions about the potential future success of the drug, highlighting the benefit of the AI methodologies’ predictive power.
"In head-to-head comparisons, we demonstrated the superiority of this approach over existing methodologies to accurately predict 'winners' and 'losers' in clinical trials," said Pradipta Ghosh, MD, one of the lead authors of the study.
The authors of the study note that next steps include further clinical development of the AI-identified drug candidate, which may shed light on whether this methodology could be useful for other diseases.