Amultidisciplinary team of researchers from the National University of Singapore (NUS) has developed an artificial intelligence (AI) technology platform that could potentially change the way drug combinations are being designed, hence enabling clinicians to quickly determine the most effective drug combination for a patient.
Applying the platform towards drug-resistant multiple myeloma, the researchers were able to establish new effective drug combinations, as well as identify the patients who may be more responsive to these treatments in under a week.
Existing methods for designing drug combinations typically involve testing arbitrary combinations of commonly used drugs or incorporating new targeted therapies into established drug combinations. Bortezomib-containing drug combinations are currently used as the first- and second-line treatment for multiple myeloma. However, most patients inevitably become resistant to these drugs and new combinations need to be established. While some newer combinations have shown to be effective for some patients, rapid identification of an optimal personalized treatment for a specific patient from an infinite span of possible drug combinations remains a challenge.
An Optimal Drug Combo
The research team comprising clinicians, engineers, and molecular biologists from NUS have developed an AI technology platform, quadratic phenotypic optimization platform (QPOP), to speed up drug combination design and identify the most effective drug combinations targeted at individual patients using small experimental datasets. With just a small amount of blood or bone marrow sample from patients, the platform is able to map the drug response that a set of drug combinations will have on the specific patient's cancer cells.
From an initial pool of 114 FDA-approved drugs, QPOP was able to identify a series of effective drug combinations, including a completely novel and unexpected combination that outperformed the standard of care regimen for relapsed myeloma. The performance of the novel combination was validated against 13 patient samples. QPOP was also used to fine-tune dosage ratios of the novel combination for optimal effectiveness.
Using four other patient samples, the research team further demonstrated that QPOP was able to evaluate and rank the novel combination against the other two current clinically used drug combinations. The novel combination was found to be the most effective treatment option for two of the myeloma patient samples tested. While the novel combination was not the most effective for all four patient samples, QPOP was able to match the best drug combination to each patient, hence demonstrating proof-of-concept for personalized medicine.
The ability of QPOP to simultaneously incorporate optimization and personalization of drug combinations with unprecedented speed opens the door to improving patient accessibility to personalized medicine. New combination therapies can also significantly broaden the spectrum of effective treatments for patients.
"QPOP revolutionizes the way in which drug combinations are designed and represents a key area in health care that can be transformed with AI," noted Edward Kai-Hua Chow, PhD, Principal Investigator at the Cancer Science Institute of Singapore, NUS, who led the study. "The efficiency of this platform in utilizing small experimental datasets enables the identification of optimal drug combinations in a timely and cost-efficient manner, which marks a big leap forward in the field of personalized medicine."
The team is now working towards translating this work into the clinic. They will look into recruiting patients for prospective clinical trials related to QPOP and other AI platforms in 2019, as well as expand the application of the platform into other disease areas.