Predicting drug flops with AI

In March 2019, Biogen’s long anticipated drug for Alzheimer’s disease aducanumab failed in two Phase III trials. While a setback, it wasn’t unusual. Drug failures are worryingly common in life sciences; approximately 90% of clinical trials fail.

Failures like this waste a significant proportion of time and resources; it has been estimated that it costs, on average, $2.6bn to develop a new drug.

Aducanumab’s failure reduced Biogen’s market value by almost a third, with shares falling by 29% overnight; this is on top of the $830m wasted by Biogen and Eisai on these trials over the past three years, according to Bloomberg.

However, Frankfurt-based Innoplexus can do what is commonly perceived in life sciences to be impossible: successfully predict the outcome of clinical trials. And in the case of aducanumab, it appears their technology was dead on.

Its artificial intelligence (AI)-based Clinical Trial Prediction (CTP) engine predicted there was between a 70% and 90% chance that Biogen’s aducanumab would fail. It was also able to pinpoint precisely why the trial failed; Innoplexus’ CEO Dr Gunjan Bhardwaj attributed this failure to some of the sites chosen, and “severely questionable” execution risk, as well as issues with the drug itself.

Although it started by back-testing completed trials, this isn’t the first time Innoplexus has successfully a predicted drug failure or success while it is ongoing; Bhardwaj claims to have an 85% precision rate. He points to Sage Therapeutic’s Zulresso (brexanolone), which succeeded in Phase III, and led to an increase in 45% of market value for the company within one day, as a notable example of a trial the company predicted would succeed.

The ability of Innoplexus’ AI tool to predict the outcome of clinical trials at any stage of the drug discovery and development progress means that it can also pinpoint the main failings in advance and help pharma companies can better manage risk fundamentally transform drug discovery and development.

Innoplexus’ algorithm can “crawl more than 2.5 billion web pages a day” and pull out relevant parameters, such as those related to “drug disease pathways…, what people who work in clinical development call execution risk, what sites are being used…, [and] what endpoints have been chosen”.

The algorithm can study around 350 parameters within the CTP engine to predict the success probability of a clinical trial; the company is planning to add further more features and parameters in the future.

Since the system is fully automated and searchable in real time, the probability of success changes and becomes more accurate as more data is integrated into the system.

To improve accuracy by increasing the amount of data the algorithms has access to, Innoplexus has linked its AI engine to blockchain with the aim of mobilising data trapped in silos in a secure way and establishing a database.

Bhardwaj believes “AI in collaboration with blockchain, is the silver bullet, specifically in the life sciences industry” as it means scientists know their data will not be misused or stolen by another party.

Innoplexus is currently in discussions with large companies about integrating their unpublished data into the database. Their data would then be combined with Innoplexus’ existing dataset, meaning the CTP’s algorithm could predict the success probability of their portfolio with greater accuracy and resolution.

Having had successful collaborations to date, Innoplexus is seeking to launch more programmes and partnerships with pharma companies and clinical trial organisations. The aim will be to establish which of the company’s trials has the highest probability success, as well as predicting the risk of in-licensing of drugs compared to continuing with an in house drug.

The impact that clinical trial failure has on the share price is a major focus for Innoplexus in developing its algorithms and prediction engine.

Therefore, Bhardwaj notes that the company wants to expand to focus on “other events that affect the market performance of life science companies”, such as regulatory events, and research and development milestones.

This will further increase CTP’s capabilities to include predicting the entire market performance of a drug and thus further mitigating risk associated with clinical development.