Gustave Roussy Institute and Paris-Sud researchers and engineers have developed an artificial intelligence system called Resolved2. The program has been designed to assess prospective cancer drugs and probability of FDA approval. If this tool is accurate, it can save drug developers enormous amounts of time and money in the aggregate.
While a growing number of cancer drugs include an increasing number of antineoplastic agents (ANAs) candidates in Phase I trials, at the same time the attrition rate for final approval grows. The French-based research team sought to overcome this growing challenge by developing a machine learning algorithm to predict drug development outcomes. This would be immensely valuable, if accurate, as it would support drug development sponsors to make earlier go/no-go decisions after Phase I clinical trials by better selecting the drugs that should proceed to Phase II.
In developing Resolved2, the team utilized PubMed abstracts of Phase I clinical trials reporting on ANAs together with pharmacologic data from the DrugBank 5.0 database to model time to U.S. Food and Drug Administration (FDA) approval) or what they title “FDA approval-free survival” since the first Phase I clinical trial publication. Resolved2 was trained, using machine learning, to understand attributes involved with progress and success vs. termination. Its performance was evaluated on an independent test set with weighted concordance index (IPCW).
After identifying 462 ANAs from PubMed matching with DrugBank 5.0 (Phase I clinical trials from 1972 to 2017), they factored in 28 variables (out of 1,411) utilized by Resolved2 to model the FDA approval-free survival—an IPCW of 0.89 on the independent test set.
Resolved2 outperformed a model that was based on efficacy/toxicity (IPCW, 0.69). In a test set at 6 years follow up, 73% (95% CI, 49% to 86%) of drugs predicted to be approved were approved, whereas 92% (95% CI, 87% to 98%) of drugs predicted to be nonapproved were still not approved (log-rank P < .001). A predicted approved drug was 16 times more likely to be approved than a predicted nonapproved drug (hazard ratio, 16.4; 95% CI, 8.40 to 32.2).
The team articulated in their summary that once the Phase I clinical trial is completed, the machine (Resolved2) can predict accurately the time to FDA approval. They offer proof-of-concept that drug development outcomes can be predicted by machine learning strategies.
The Research Centers
Gustave Roussy is a cancer-research institute and European Cancer Center. It is a center for patient care, research and teaching—and patients with all types of cancer can be treated there. It is located in Villejuif, South Paris, France. It is named after the Swiss-French Neuropathologist Gustave Roussy. The center has 457 beds and was founded in 1926.
Paris Sud University is a French research university distributed among several campuses in the southern suburbs of Paris including Orsay, Cachan, Châtenay-Malabry, Sceaux and Kremlin-Bicêtre campuses. Founded in 1971, they run on a yearly budget of approximately 450m Euros, employ 2,461 academic staff and about 27,000 students.
Call to Action: If Resolved2 is, in fact, accurate, it could save drug developers enormous amounts of time and money. But there are many assumptions and details here that need to be sorted out. The tool should be further tested and verified. It could be a very valuable tool.