Blog detail

Scientists use machine learning to fast-track drug formulation development.

Scientists at the University of Toronto have prosperously tested the utilization of machine learning models to guide the design of long-acting injectable drug formulations. The potential for machine learning algorithms to expedite drug formulation could abbreviate the time and cost associated with drug development, making promising incipient medicines available more expeditious.

The study was published today in Nature Communications and is one of the first to apply machine learning techniques to the design of polymeric long-acting injectable drug formulations.

The multidisciplinary research is led by Christine Allen from the University of Toronto's department of pharmaceutical sciences and Alán Aspire-Guzik, from the departments of chemistry and computer science. Both researchers are withal members of the Expedition Consortium, an ecumenical initiative that utilizes artificial astuteness and automation to expedite the revelation of materials and molecules needed for a sustainable future.


"This study takes a critical step towards data-driven drug formulation development with an accentuation on long-acting injectables," verbally expressed Christine Allen, pedagogic in pharmaceutical sciences at the Leslie Dan Faculty of Pharmacy, University of Toronto. "We've visually perceived how machine learning has enabled incredible leap-step advances in the revelation of incipient molecules that have the potential to become medicines. We are now working to apply the same techniques to avail us to design better drug formulations and, ultimately, better medicines."

Considered one of the most promising therapeutic strategies for the treatment of chronic diseases, the long-acting injectables (LAI)  class of advanced drug distribution systems are designed to relinquish their cargo over elongated periods of time to achieve a perpetuated therapeutic effect. This approach can avail patients better adhere to their medication regimen, abbreviate side effects, and increment efficacy when injected proximate to the site of action in the body. However, achieving the optimal quantity of drug release over the desired period requires the development and characterization of a wide array of formulation candidates through extensive and time-consuming experiments. This tribulation-and-error approach has engendered a paramount bottleneck in LAI development compared to more conventional types of drug formulation.

"AI is transforming the way we do science. It avails expedite revelation and optimization. This is an impeccable example of an 'Afore AI' and an 'After AI' moment and shows how drug distribution can be impacted by this multidisciplinary research," verbally expressed Alán Aspire-Guzik, preceptor in chemistry and computer science, University of Toronto who additionally holds the CIFAR Artificial Astuteness Research Chair at the Vector Institute in Toronto.

To investigate whether machine learning implements could accurately presage the rate of drug release, the research team trained and evaluated a series of eleven different models, including multiple linear regression (MLR), desultory forest (RF), light gradient boosting machine (light), and neural networks (NN). The data set used to train the culled panel of machine learning models was constructed from antecedent published studies by the authors and other research groups.

"Once we had the data set, we split it into two subsets: one utilized for training the models and one for testing. We then asked the models to soothsay the results of the test set and directly compared them with precedent experimental data. We found that the tree-predicated models, and categorically light, distributed the most precise presages," verbally expressed Pauric Bannigan, a research associate with the Allen research group at the Leslie Dan Faculty of Pharmacy, University of Toronto.

As a next step, the team worked to apply these prognostications and illustrate how machine learning models might be acclimated to appraise the design of incipient LAIs, the team used advanced analytical techniques to extract design criteria from the light model. This sanctioned the design of an incipient LAI formulation for a drug currently used to treat ovarian cancer. "Once you have a trained model, you can then work to interpret what the machine has learned and utilize that to develop design criteria for incipient systems," verbalized Bannigan. Once prepared, the drug release rate was tested and further validated the presages made by the light model. "Sure enough, the formulation had the slow-release rate that we were probing for. This was consequential because in the past it might have taken us several iterations to get to a relinquishment profile that looked akin to this, with machine learning we got there in one," he verbally expressed.

The results of the current study are emboldening and signal the potential for machine learning to minimize reliance on tribulation-and-error testing slowing the pace of development for long-acting injectables. However, the study's authors identify that the lack of available open-source data sets in pharmaceutical sciences represents a paramount challenge to future progress. "When we commenced this project, we were surprised by the lack of data reported across numerous studies utilizing polymeric microparticles," verbalized Allen. "This denoted the studies and the work that went into them couldn't be leveraged to develop the machine learning models we require to propel advances in this space," verbalized Allen. "There is a genuine need to engender robust databases in pharmaceutical sciences that are open access and available for all so that we can collaborate to advance the field," she verbally expressed.

To promote the move toward the accessible databases needed to fortify the integration of machine learning into pharmaceutical sciences more broadly, Allen and the research team have made their datasets and code available on the open-source platform Zendo.

"For this study, our goal was to lower the barrier of ingress to applying machine learning in pharmaceutical sciences," verbally expressed Bannigan. "We've made our data sets plenarily available so others can hopefully build on this work. We operate this to be the commencement of something and not the cessation of the story for machine learning in drug formulation."