An article recently published in Genetic Engineering & Biotechnology News, heralds DrugCell, an experimental artificial intelligence (AI) system.
DrugCell was created at the Moores Cancer Center and the School of Medicine at the San Diego University with the ultimate goal of identifying the most effective drug treatments for malignant tumors. The team inputs the tumor’s data into the system, and DrugCell identifies the most appropriate treatment. Identification includes the best drug, or combination of drugs, and the pathways that may be controlling drug response.
Unimpressive Statistics
Out of all the therapeutic drugs for cancer that are submitted for FDA approval, only a meager four percent receive final approval.
Dr. Trey Ideker at Moores Cancer Center acknowledges that currently, scientists are finding it difficult to find the best drug combinations. This is especially true for cancer drugs. The doctor explained that scientists have not been able to identify the best drug combinations because of the complexity of tumor cells and the lack of a complete understanding of drug response.
Although the techniques used in machine learning will eventually create the coveted drug response predictions, they have not yet entered the clinical practice arena.
Dr. Ideker said that in order for AI to be effective, scientists must learn how the system generates its conclusions. In other words, why a particular decision was made, the pathways the drugs are targeting, and the reason behind a drug’s rejection or response.
About Machine Learning Techniques
Dr. Ideker and his team developed DrugCell, which involves a deep learning, decipherable model of cancer cells. The cells were cultured based on responses from twelve hundred tumor cell lines. The AI system identified six hundred eighty-four paired drugs.
The mechanism behind DrugCell allows the team to input the tumor’s data. The system will identify:
· The most effective drug
· The best combination of drugs to treat the cancer
· The pathways that produce a controlled response to that specific drug
DCell, the First AI System
The team initially created DCell, an AI system that used information taken from a yeast cell’s mutations and genes. DCell was able to predict a cell’s growth and how a cell would act.
DrugCell followed DCell as the next-generation model. DrugCell focused on twelve hundred tumor cell lines (perpetuating strains of cells) together with their responses to almost seven hundred FDA-approved therapeutic drugs.
This amounts to over five hundred thousand drug matches. Also, in laboratory experiments, the team was able to confirm a list of DrugCell’s predictive responses.
About Tumor Genotypes
Tumor genotypes (collection of genes) enable a cell’s subsystem to predict treatment response and understand the mechanisms behind drug response.
When analyzing DrugCell’s mechanisms, a scientist is led to drug combinations that can be validated by:
· CRISPR gene-editing technology
· drug-drug testing in-vitro (in test tubes)
· patient-derived xenografts which are tumor tissue that, for research purposes, has been taken from a patient and implanted into mice.
It can be said that DrugCell creates a blueprint for the construction of definable models for predictive medicine.
The DrugCell team can input the tumor’s data, and the system will return the most effective drug, the biochemical pathways (interactions between genes) controlling response to the drug, and the best combination of drugs that will treat the cancer.
Precision Cancer Therapy
Moores Cancer Center has made precision therapy available at UCSD Health. Patients can have their tumor biopsied, sequenced, and assessed by an interdisciplinary group of doctors and at the Molecular Tumor Board.
A recent study found that patients who take part in personalized treatment will experience better results. It can be said that DrugCell acts as the Molecular Tumor board.
The Ultimate Goal
The DrugCell team readily admits that there is still a lot to accomplish in order to reach their ultimate goal of getting DrugCell into clinics.
They understand that twelve hundred cell lines do not fully account for cancer’s diversity. The team will be adding additional data and experimenting with various drug structures. They also have plans to form partnerships with active clinical trials in order to embed DrugCell and test it in the “real world”.
