Identifying biopharmaceuticals with the potential to be formulated as pills just became easier, according to researchers who used artificial intelligence to find drugs stable enough for oral delivery.
Oral delivery is a major research focus for formulators. Demand for needle-free, compliance-boosting delivery methods for biopharmaceuticals has seen R&D labs the world over checking whether products can be turned into tablets. But developing oral biopharmaceuticals is no easy task, says Abdul Basit, PhD, a professor from University College London’s (UCL) School of Pharmacy.
“To have the desired therapeutic effect biopharmaceuticals need to reach their target in their original, intended form. Administering medicines orally is patients’ preferred administration route,” he explains. “However, biopharmaceuticals are often easily degraded in the GI tract. “[This] will likely mean that the biopharmaceutical is degraded before it can exert therapeutic activity at its target. This is why most biopharmaceuticals are currently administered by injection.”
The big challenge is identifying those large molecule drugs with the characteristics to survive the rigors of the gastrointestinal tract, continues Basit, because current methods are complex, time consuming, and rely on animal models.
“Biotech companies will conduct a range of experiments to ensure that their biopharmaceutical is not rapidly degraded in relevant environments, such as blood or GI fluids. This could involve incubating the drugs with either the real fluids, but more commonly simulated or animal fluids, and measuring drug concentration over a certain time period,” he tells GEN.
“When the biopharmaceutical is administered to animals and patients then researchers can measure the amount of intact drug in relevant body fluids, such as blood, urine, or stool. This can give an indication of where and how the drug is degraded.”
To try and address this, Basit and UCL colleagues developed an in silico way of predicting peptide stability. The approach—detailed in a paper published in the International Journal of Pharmaceutics—uses machine learning (ML) to speed up the process significantly.
“We chose ML to predict the GI stability of peptides because currently researchers must test every one of their drug candidates in the lab, which can be expensive and time- and resource-consuming,” he points out. “We wanted to give researchers the opportunity to rapidly screen new peptide-based drugs for suitability for oral administration. Thus, they can potentially save time by experimentally testing only the most promising candidates.
“We picked ML as a predictive technology rather than, for example, design of experiments or molecular modeling as it is flexible, works well with large datasets, and can model data with complex relationships. Further, it is rapid, accessible, and leverages the current evidence in the literature to enable predictions for untested peptide drugs.”
In fact, the models are already being used. UCL spin-out Intract Pharma has already applied them to its pipeline of biopharmaceutical candidates.
And, while the focus was on predicting stability in the GI tract, the ML techniques used could be further developed for use in manufacturing, according to Basit.
“While the features that confer GI stability could potentially be similar to those that confer manufacturing stability, prediction of manufacturing stability is a different task and thus requires a different model,” he notes. “Future work could explore whether an emerging form of ML, known as transfer learning, could transfer knowledge pertaining to GI stability to manufacturing stability.”