Machine learning to de-risk clinical trials

Insilico and Oxford University used machine learning to predict major clinical forms of drug cardiotoxicity.

Insilico Medicine announced this week that they, in partnership with the Computational Cardiovascular Science Group of the University of Oxford, have overcome one of the hurdles of late-stage clinical trial failures and predicted a range of drug-related adverse cardiac outcomes.

Longevity.Technology: It is frustrating when clinical trials fail at late stage due to something as damning as toxicity. This research shows that toxicity can be predicted, paving the way for the prevention of expensive failures and earlier safety evaluation as well as improving the productivity of drug discovery pipelines.

The research collaboration demonstrated the simultaneous prediction of cardiotoxic relationships for six drug-induced cardiotoxicity types. The team used a machine learning approach on a large collected and curated dataset of transcriptional and molecular profiles.

The team used machine learning to investigate 1,131 drugs, 35% with known cardiotoxicities, looking at 9,933 samples in all. Their best prediction attained an average accuracy of 79% for safe versus risky drugs, across all six cardiotoxicity types and an average accuracy of 66% on the unseen set of drugs.

The results also showed that individual cardiotoxicities for specific drug types were also predicted with high accuracy, including cardiac disorder signs and symptoms for a previously unseen set of anti-inflammatory agents, as well as heart failures for an unseen set of anti-neoplastic agents [1].

Leading author Professor Blanca Rodriguez said: “Drug-induced adverse effects on the heart are a very important problem … we are very excited to show how our machine learning algorithm can identify drugs that can cause 6 potential forms of cardiac adverse outcomes from gene expression data [2].”

“…increased chances for translation into drug discovery and development pipelines … a valuable benchmark for future studies.”

Polina Mamoshina is a Senior Research Scientist at Insilico Medicine. Commenting on the research, she said: “In silico or computational models have made great progress in past years.

And one of their great features is that they can be humanized and so have increased chances for translation into drug discovery and development pipelines. The scope of this work was to predict drug adverse reactions that were shown in humans. We believe that this work can be extended to side effects manifested in other organs and tissues and that pipeline that we proposed provides a valuable benchmark for future studies [3].”

Alex Zhavoronkov, founder and CEO of Insilico Medicine, added: “Drug-induced cardiotoxicity is one of the reasons for late-stage clinical trial failures. We see the Rodriguez group at Oxford as the world’s main source of accurate cardiotoxicity predictors.

The results of their work are adopted by the FDA, and many pharmaceutical companies [4].”
[3] Ibid
[4] Ibid

Image of Polina Mamoshina courtesy of Insilico Medicine