A new AI model can accurately predict human response to novel drug compounds

The FDA approval of a new drug can take a decade and cost a billion dollars. An artificial intelligence model created by a research team at the CUNY Graduate Center could greatly improve the accuracy and reduce the time and cost of drug development.

The new model, called CODE-AE, was described in a paper in Nature Machine Intelligence. It was able to identify personalized drugs for thousands of patients. The technique is expected to speed up drug discovery.

Predicting patient-specific responses to a new chemical compound is crucial to discovering safe and effective therapies and to selecting an existing drug for a specific patient. It's not feasible to do early efficacy testing of a drug in humans. Tissue models can be used to evaluate the therapeutic effect of a drug molecule. The drug effect in a disease model doesn't correlate with the drug efficacy and toxicity in humans. High costs and low productivity rates of drug discovery are caused by a knowledge gap.

The senior author of the paper said that the new machine learning model can address the challenge ofTranslation from disease models to humans. CODE-AE takes advantage of recent advances in machine learning. One of its components uses the same techniques.

The problem of having enough patient data to train a machine learning model can be solved by the new model. Many methods to use cell-line screens for predicting clinical responses are unreliable due to data incongruity and discrepancies. The data-discrepancy problem can be alleviated with the help ofCODE-AE.

CODE-AE significantly improves accuracy and robustness over state-of-the-art methods in predicting patient specific drug responses.

The next challenge for the research team is to develop a way to reliably predict the effect of a new drug on human bodies. The researchers said that the model could be adjusted to predict human side effects to drugs.

More information: A Context-aware Deconfounding Autoencoder for Robust Prediction of Personalized Clinical Drug Response From Cell Line Compound Screening, Nature Machine Intelligence (2022). DOI: 10.1038/s42256-022-00541-0 Journal information: Nature Machine Intelligence