It’s never been a chip shot developing cancer treatments. In fact, only 7.78% of all cancer drugs that enter phase 1 clinical trials are ultimately approved by the U.S. Food and Drug Administration (FDA), according to the Pharmaceutical Manufacturers Association. Of the 1,481 cancer drugs selected for clinical trials between 2000 and 2020, only 115 were approved. In a world where that process costs nearly $650 million on average for a single cancer drug, a 92.22% rate of failure means billions of dollars and thousands of hours wasted.
And the number of cancer drugs entering pre-clinical and clinical trials is on the rise. In 2020 23.1% of all clinical trials are oncology drug trials; that is up from 14.8% in 2000, according to ClinTrials.gov. In 2021, 1,300 drugs had been selected for pre-clinical and clinical trials. Without a change in approach, companies will continue wasting gargantuan amounts of resources in hopes they find that rare diamond in the rough.
Why is it that so many cancer drugs fail in clinical trials? And how can modern technology like artificial intelligence, 3D cell cultures, and novel approaches to tumor research improve the odds? After more than a century of research, we’re finally on the cusp of improving our ability to effectively discover and develop safe, effective cancer drugs not only for specific types of cancer, but also specific types of patients.
Why most cancer drugs fail to attain approval
Cancer is an extremely complex disease that evolves and changes as it grows. This makes it nearly impossible to create a one-size-fits-all cancer treatment that has a high rate of success across a diverse patient population.
A new drug candidate that begins with a promising study in laboratory cell testing can quickly move into more expensive animal trials as companies try to advance their most promising projects. Even if it performs well in animal trials, that same drug can fizzle once it enters human trials (after significant investments in time and money have already been made).
Moreover, even when a drug shows success in treating one patient demographic means nothing when it comes to treating patients with entirely different genetic backgrounds. Physicians from around the world have always known the drugs that work on a population with a genetic history based in Europe may have reduced efficacy or not even work at all for patients with genetic profiles rooted in the Middle East North Africa (MENA) region.
Similarly, drugs used in traditional medicines in Eastern cultures often have no effect on Europeans and North Americans, except for those communities that originated from the East. In some cases, these drugs can even be toxic to patients with certain genetic profiles.
Adding to this built in diversity or heterogeneity of the people with cancer, the cancer itself becomes highly diverse or heterogeneous. Dr. Mary Relling, Ph.D., and her team at St. Jude’s Hospital have identified a number of Single Polynucleotide Polymorphisms (SNPs) called ‘snips’, or changes in the DNA code amongst different adolescent patients, that shows the bad and good effects of having ‘inherited’ polymorphisms in your genetic code. These polymorphisms can cause a child to respond favorably or unfavorably to combination drug treatments for cancer.
Needless to say, all this makes it very difficult to predict treatment outcomes when it comes to cancer. In order to do so, researchers would need to take each of the following into account.
As cancer grows, tumor heterogeneity tends to increase, which means the tumor becomes more complex and contains a wider range of different cell types. As a result, it becomes more difficult to treat cancer as it progresses; the more complex it becomes, the more cancer confounds oncologists and evades treatment plans.
Tan Ince, MD, Ph.D., now Chief of Pathology at Cornell-Weill NewYork-Presbyterian Brooklyn Methodist Hospital, wasn’t satisfied with the existing ways of culturing tumor cells in the pathology lab to screen drugs against patient samples. While working at Brigham and Women’s as a pathologist and serving at Robert Weinberg’s (Ph.D.) Laboratory at MIT’s Whitehead Institute, Dr. Ince challenged this model on grounds that it didn’t accurately account for tumor heterogeneity or the dynamic nature of cancer as tumors grow.
Ince’s work helped the Weinberg lab identify and openly challenge the screening of drugs using outdated laboratory procedures for culturing cells that were developed over 60 years ago. Dr. Weinberg also rightly points out in his lectures and papers, that “mice aren’t people” and we should stop selecting drugs based on any models that don’t accurately reflect the human cancer from a patient.
In one of his papers on ovarian cancer, Ince also developed a media that can be used to grow ovarian cancer cells in a lab while retaining the signature of the tumor or cancer that is in the patient. The media Ince developed is available to all researchers around the world today, so no laboratory should be using old methods that are not as effective. Using Ince’s media, researchers can grow ovarian cancer cells and test drugs against a true patient sample to see which combination of drugs will really work for treating the cancer as it exists in the patient’s body.
Even the same type of cancer can affect two different patients in very different ways. This is because genetic heterogeneity is just as important to devising an effective cancer treatment as understanding tumor heterogeneity.
For example, patients of eastern European descent are susceptible to the BRCA1 and BRCA2 genetic mutations, which could lead to the development of breast cancer and ovarian cancer. Treating these patients, then, would likely require different drug formulations than treating breast cancer or ovarian cancer patients who have a sub-Saharan African genetic background, for example, as Relling has shown in adolescents – the same is true in older patients as well.
A good example of how even newly approved drugs suddenly stop working – or how the patient develops resistance to treatment – is found in the newly developed PARP inhibitors that were specifically designed for the BRCA1 and BRCA2 patients. These PARP inhibitors work remarkably well – until they don’t. Dr. Stephen Taylor from the University of Manchester has shown that PARP inhibitors and PARG inhibitors can work together effectively in certain ovarian cancer patient populations of BRCA1 and BRCA2 patient sets.
But in order to know when they will be effective, oncologists and researchers must be able to predict treatment outcomes based on all the factors tumor heterogeneity, all the factors that go into genetic heterogeneity, and all the unique lifestyle and environmental factors that influence each individual patient.
It’s just too much information to contextualize — until now. At Predictive, we bring the patient into the very heart of drug discovery and development.
Using artificial intelligence and machine learning to improve cancer treatments
The issue with cancer research historically is that it simply isn’t feasible, even for teams of experts, to possibly account for the many factors that influence how a tumor develops and grows in a particular patient and then account for which drug formulations will best treat it. That’s because the human mind can only handle so much stimuli.
Machine learning and artificial intelligence can process information so many orders of magnitude faster than human beings that it can complete countless scenarios, predicting cancer treatment outcomes with various different drug formulations, before a team of human experts has played out even a single scenario. And it can do that around the clock, 24/7/365, getting sharper and smarter as it goes. AI can improve cancer research by rapidly considering more factors about the cancer, the patient, and the treatment than humanly possible.
That’s precisely what Predictive Oncology is doing with its team at Helomics, applying three proprietary machine learning algorithms to a massive database of more than 150,000 de-identified patients, 131 types of tumors, and 30 different types of cancers. These algorithms — known as CoRETM, PeDALTM, and TruTumor™ — are able to analyze all this data to determine:
- the top optimal drug formulations
- for the individual patient based on their genetic background and lifestyle
- based on the type of cancer and heterogeneous make-up of the tumor in its current stage.
Armed with this information, sussed out through millions of scenarios analyzed by the AI, pharmaceutical companies would no longer have to waste time and resources on a wild goose chase developing drug formulations that were never going to pass human clinical trials in the first place. Instead, the targeted development of safe and effective drug formulations for specific types of cancer in clearly identified patient populations means a faster discovery process, streamlined drug development, and a clear path to U.S. Food and Drug Administration (FDA) approval.
AI’s lifesaving potential and remaking the healthcare industry
AI and healthcare are a natural fit, enabling oncologists and pharmaceutical manufacturers to create more effective treatment plans with more useful drug formulations. But it’s not only the cancer research space that AI is set to disrupt — machine learning algorithms and their ability to contextualize immense troves of patient data are poised to remake the healthcare industry as a whole.
Market research projections from industry analysts at Reports and Data anticipate that healthcare AI will grow to $61.59 billion in value by 2027, a compound annual growth rate (CAGR) of 43.6%. And we’ve only scratched the surface of what AI can do for human health and wellness. The lives that will be saved and the value that will be created thanks to these technologies and the novel approaches to healthcare research they enable will be immeasurable. And with the dawn of the AI-powered healthcare future, there also comes hope for a day when we finally eliminate cancer.