How Can AI Help Predict Patient Drug Response? (Genialis)
Computational biology and AI are already used to expedite the creation of more precise medications. Unfortunately, this still has limited impact on expediting clinical trials, due to the challenges that persist in recruiting patients.
Future promises lie in digital twins, that could act as a digital human placebo arm, potential exclude animal models and prevent patient harm, since trial medications would be tested through computational models, before given to people, says Rafael Rosengarten, CEO of Genialis.
Drug development has always been and remains expensive. Estimates are conflicting. The long-cited number to bring a drug to market has been 3 billion US dollars. A 2020 study corrected that estimate to be between $985 million and $1.3 billion per new drug. Latest research raises the estimate to whooping $6 Billion.
Needless to say, we could use optimization, since new drugs are essential for improved care and quality of life of patients.
Various companies have for years used large data sets and AI to address various aspects of drug development: matching patients to clinical trials, discovering new targeted molecules, or predicting patient drug response.
Genialis uses machine learning and omics data to predict patient responses to targeted therapies based on underlying disease biology. Their computational models consider a wide array of evidence, including RNA sequencing data, to understand diseases at a deeper level. These models go beyond simple targeted therapy and provide informative insights into disease phenotypes for better treatment decisions.
AI and Clinical Development
So why is the impact of AI still limited? The clinical development timelines are slow and expensive. “We're seeing evidence already of the speed with which you can discover targets and discover new molecules, right? So the early stage stuff is moving much faster already, the kind of preclinical validation is the evidence you need to get to an IND (investigational new drug). But in clinical development, AI hasn’t started to impact the timelines as much as we might hope. One of them is just, recruitment is hard, right? So when you have 10,000 clinical trials running for PDL-1 combinations, how are you gonna find enough patients for all those trials? It's hard to recruit, hard to enroll. Access is a huge issue. We know that the gulf in information between big cancer centers and community hospitals and community clinics is huge,” explains Rafael Rosengarten.
Precision Medicine and Pricing
Efforts are being made to use AI in areas such as matching patients to trials and ranking trial sites for better efficiency. Precision medicine is becoming more important, but large-scale studies may not always be feasible. Digital twins, where a patient's digital avatar serves as a control group, could help address this issue. Drug pricing is also a concern, but decreasing drug development time and improving success rates can help reduce costs.
The future of healthcare will likely involve policymakers and providers insisting on outcome-based care. Bias in healthcare data sets is a significant problem that needs to be addressed through inclusive data curation and validation processes. Generative AI has had an impact on discovering new molecules in pharmaceuticals, while generative biology can lead to new targets and biomarkers. Overall, AI has made progress in the last decade but still faces challenges related to clinical development timelines and bias in data sets.
Drug repurposing and indication expansion are seen as immediate priorities in leveraging existing drugs for new treatments, says Rafael Rosengarten. While discovering new drugs is important but time-consuming, repurposing existing drugs could lead to faster results.
Companies also face data challenges, especially operating in different markets. Restrictions can sometimes be overcome by using federated learning solutions or working with data on-site when it cannot be exported.
The most significant challenge faced by Genialis is the availability of high-quality data for their solution development. They curate fit-for-purpose datasets through partnerships and extensive research efforts. Building these datasets takes time, energy, and careful selection of patient samples.
This is just an excerpt. Tune in to the full episode.