To say that identifying successful drug candidates during development is the 64,000-dollar question is a huge understatement!
Much more money – millions, tens, or even hundreds of millions of dollars – rides on picking the winner: the drug that beats the odds and eventually gets approved by the FDA, the EMA, or another regulatory body.
Assessing and adjusting the probability of success of a compound as it moves on from pre-clinical to and through the increasingly expensive clinical phases has been notoriously difficult. A fact that is proven by a sobering statistic: less than 1 in 10 drugs that enter clinical development will eventually be commercialized.
Drug developers have established approaches to evaluate the Probability of Technical and Regulatory Success (PTRS) of a drug in development to help them make informed decisions, but the low drug approval rates show that there is significant room to improve these risk assessments.
Let’s take a quick look at how PTRS is assessed currently and why artificial intelligence, specifically machine learning (ML), is such a powerful tool for improving risk assessments.
Current approach: historical data and experience
While each pharmaceutical company has its own approach to coming up with PTRS scores, they tend to all incorporate the following information:
- Historical estimates based on the current development phase of the program and the specific disease. These estimates serve as the baseline that then gets adjusted with the following information
- Input from their external experts or key opinion leaders, mostly experienced physicians and drug developers.
- Statistical analysis performed in-house by the R&D analytics groups that take parameters such as available clinical data, drug mechanism of action, and clinical trial design into account.
The resulting PTRS scores are, therefore, based on limited data and are rather heavily influenced by the experience of a few key individuals.
AI/ML – the perfect tool for the job
Crunching a lot of data and learning to “see” trends and patterns is exactly what AI/ML excels at which makes it the perfect tool.
We have decades’ worth of detailed information about many aspects of drug development available in databases. Information such as clinical trial design, clinical trial outcomes, regulatory data, information about drug biology as well as about the companies sponsoring the trials can be pulled, curated, and then serve as a large and objective foundation on which to base PTRS assessments.
In addition, we have modern ML algorithms which are basically pattern recognition machines. They operate similarly to humans: they learn by “seeing” and that learning enables them to identify patterns and detect connections that were previously hidden. While humans are very good at pattern recognition in their own right, the amount of data that needs to be considered for PTRS assessments far exceeds human processing power.
The vast amount of available data and rapid innovations in machine learning means that AI-based assessment of the probability of success of a drug candidate is a promising approach.
At Intelligencia, this is what we are doing. We have spent the last 5 years curating data from a multitude of different data sources as well as selecting, honing, and training a variety of different algorithms.
A glance at our insights
The obvious next question is “How well do your algorithms perform? Let’s see some data!”
Here you go with a glance at our insights:
At the end of 2021, Intelligencia provided predictions on all Oncology programs. By the end of 2022, we had correctly predicted the majority of them
Of the 37 oncology clinical programs that received FDA approval in 2022, we correctly predicted 75%. Of the 898 programs that failed across all phases of development, we correctly predicted 77%.
There is obviously much more insight to be gleaned from our data. If you want to learn more, contact us to set up a call with a member of our team.